Upload 28 files
Browse files- .gitignore +1 -0
- README.md +2 -0
- app.py +58 -0
- diffusion_module/__pycache__/nn.cpython-39.pyc +0 -0
- diffusion_module/__pycache__/unet.cpython-39.pyc +0 -0
- diffusion_module/__pycache__/unet_2d_blocks.cpython-39.pyc +0 -0
- diffusion_module/__pycache__/unet_2d_sdm.cpython-39.pyc +0 -0
- diffusion_module/nn.py +183 -0
- diffusion_module/unet.py +1315 -0
- diffusion_module/unet_2d_blocks.py +0 -0
- diffusion_module/unet_2d_sdm.py +357 -0
- diffusion_module/utils/LSDMPipeline_expandDataset.py +179 -0
- diffusion_module/utils/Pipline.py +361 -0
- diffusion_module/utils/__pycache__/LSDMPipeline_expandDataset.cpython-39.pyc +0 -0
- diffusion_module/utils/__pycache__/Pipline.cpython-310.pyc +0 -0
- diffusion_module/utils/__pycache__/Pipline.cpython-39.pyc +0 -0
- diffusion_module/utils/__pycache__/loss.cpython-39.pyc +0 -0
- diffusion_module/utils/loss.py +149 -0
- diffusion_module/utils/noise_sampler.py +16 -0
- diffusion_module/utils/scheduler_factory.py +300 -0
- evolution.py +102 -0
- figs/4.png +0 -0
- figs/4_1.jpg +0 -0
- figs/4_1.png +0 -0
- figs/4_1_mask.png +0 -0
- generate.py +106 -0
- requirements.txt +137 -0
.gitignore
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
pretrain_weights/
|
README.md
CHANGED
@@ -1,3 +1,5 @@
|
|
|
|
|
|
1 |
---
|
2 |
title: LSDM
|
3 |
emoji: 💻
|
|
|
1 |
+
-- LSDM for Crack Segmentation dataset expending
|
2 |
+
|
3 |
---
|
4 |
title: LSDM
|
5 |
emoji: 💻
|
app.py
ADDED
@@ -0,0 +1,58 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
from evolution import random_walk
|
3 |
+
from generate import generate
|
4 |
+
|
5 |
+
def process_random_walk(img):
|
6 |
+
img1, _ = random_walk(img)
|
7 |
+
|
8 |
+
return img1
|
9 |
+
|
10 |
+
def process_first_generation(img1, model_path="pretrain_weights/b2m/unet_ema"):
|
11 |
+
|
12 |
+
generated_images = generate(img1, model_path)
|
13 |
+
return generated_images[0]
|
14 |
+
|
15 |
+
def process_second_generation(img1, model_path="pretrain_weights/m2i/unet_ema"):
|
16 |
+
|
17 |
+
generated_images = generate(img1, model_path)
|
18 |
+
return generated_images[0]
|
19 |
+
|
20 |
+
# 创建 Gradio 接口
|
21 |
+
with gr.Blocks() as app:
|
22 |
+
with gr.Row():
|
23 |
+
with gr.Column():
|
24 |
+
input_image = gr.Image(value="figs/4.png", image_mode='L', type='numpy', label="Upload Grayscale Image")
|
25 |
+
|
26 |
+
process_button_1 = gr.Button("1. Process Evolution")
|
27 |
+
|
28 |
+
with gr.Column():
|
29 |
+
output_image_1 = gr.Image(value="figs/4_1.png", image_mode='L', type="numpy", label="After Evolution Image",sources=[])
|
30 |
+
process_button_2 = gr.Button("2. Generate Masks")
|
31 |
+
|
32 |
+
with gr.Row():
|
33 |
+
with gr.Column():
|
34 |
+
output_image_3 = gr.Image(value="figs/4_1_mask.png", image_mode='L', type="numpy", label="Generated Mask Image",sources=[])
|
35 |
+
process_button_3 = gr.Button("3. Generate Images")
|
36 |
+
with gr.Column():
|
37 |
+
output_image_5 = gr.Image(value="figs/4_1.jpg", type="numpy", image_mode='RGB', label="Final Generated Image 1",sources=[])
|
38 |
+
|
39 |
+
|
40 |
+
process_button_1.click(
|
41 |
+
process_random_walk,
|
42 |
+
inputs=[input_image],
|
43 |
+
outputs=[output_image_1]
|
44 |
+
)
|
45 |
+
|
46 |
+
process_button_2.click(
|
47 |
+
process_first_generation,
|
48 |
+
inputs=[output_image_1],
|
49 |
+
outputs=[output_image_3]
|
50 |
+
)
|
51 |
+
|
52 |
+
process_button_3.click(
|
53 |
+
process_second_generation,
|
54 |
+
inputs=[output_image_3],
|
55 |
+
outputs=[output_image_5]
|
56 |
+
)
|
57 |
+
|
58 |
+
app.launch()
|
diffusion_module/__pycache__/nn.cpython-39.pyc
ADDED
Binary file (6.26 kB). View file
|
|
diffusion_module/__pycache__/unet.cpython-39.pyc
ADDED
Binary file (29.6 kB). View file
|
|
diffusion_module/__pycache__/unet_2d_blocks.cpython-39.pyc
ADDED
Binary file (57.3 kB). View file
|
|
diffusion_module/__pycache__/unet_2d_sdm.cpython-39.pyc
ADDED
Binary file (10.7 kB). View file
|
|
diffusion_module/nn.py
ADDED
@@ -0,0 +1,183 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""
|
2 |
+
Various utilities for neural networks.
|
3 |
+
"""
|
4 |
+
|
5 |
+
import math
|
6 |
+
|
7 |
+
import torch as th
|
8 |
+
import torch.nn as nn
|
9 |
+
|
10 |
+
|
11 |
+
def convert_module_to_f16(l):
|
12 |
+
"""
|
13 |
+
Convert primitive modules to float16.
|
14 |
+
"""
|
15 |
+
if isinstance(l, (nn.Conv1d, nn.Conv2d, nn.Conv3d)):
|
16 |
+
l.weight.data = l.weight.data.half()
|
17 |
+
if l.bias is not None:
|
18 |
+
l.bias.data = l.bias.data.half()
|
19 |
+
|
20 |
+
# PyTorch 1.7 has SiLU, but we support PyTorch 1.5.
|
21 |
+
class SiLU(nn.Module):
|
22 |
+
def forward(self, x):
|
23 |
+
return x * th.sigmoid(x)
|
24 |
+
|
25 |
+
|
26 |
+
class GroupNorm32(nn.GroupNorm):
|
27 |
+
def forward(self, x):
|
28 |
+
#print(x.float().dtype)
|
29 |
+
return super().forward(x).type(x.dtype)
|
30 |
+
|
31 |
+
|
32 |
+
def conv_nd(dims, *args, **kwargs):
|
33 |
+
"""
|
34 |
+
Create a 1D, 2D, or 3D convolution module.
|
35 |
+
"""
|
36 |
+
if dims == 1:
|
37 |
+
return nn.Conv1d(*args, **kwargs)
|
38 |
+
elif dims == 2:
|
39 |
+
return nn.Conv2d(*args, **kwargs)
|
40 |
+
elif dims == 3:
|
41 |
+
return nn.Conv3d(*args, **kwargs)
|
42 |
+
raise ValueError(f"unsupported dimensions: {dims}")
|
43 |
+
|
44 |
+
|
45 |
+
def linear(*args, **kwargs):
|
46 |
+
"""
|
47 |
+
Create a linear module.
|
48 |
+
"""
|
49 |
+
return nn.Linear(*args, **kwargs)
|
50 |
+
|
51 |
+
|
52 |
+
def avg_pool_nd(dims, *args, **kwargs):
|
53 |
+
"""
|
54 |
+
Create a 1D, 2D, or 3D average pooling module.
|
55 |
+
"""
|
56 |
+
if dims == 1:
|
57 |
+
return nn.AvgPool1d(*args, **kwargs)
|
58 |
+
elif dims == 2:
|
59 |
+
return nn.AvgPool2d(*args, **kwargs)
|
60 |
+
elif dims == 3:
|
61 |
+
return nn.AvgPool3d(*args, **kwargs)
|
62 |
+
raise ValueError(f"unsupported dimensions: {dims}")
|
63 |
+
|
64 |
+
|
65 |
+
def update_ema(target_params, source_params, rate=0.99):
|
66 |
+
"""
|
67 |
+
Update target parameters to be closer to those of source parameters using
|
68 |
+
an exponential moving average.
|
69 |
+
|
70 |
+
:param target_params: the target parameter sequence.
|
71 |
+
:param source_params: the source parameter sequence.
|
72 |
+
:param rate: the EMA rate (closer to 1 means slower).
|
73 |
+
"""
|
74 |
+
for targ, src in zip(target_params, source_params):
|
75 |
+
targ.detach().mul_(rate).add_(src, alpha=1 - rate)
|
76 |
+
|
77 |
+
|
78 |
+
def zero_module(module):
|
79 |
+
"""
|
80 |
+
Zero out the parameters of a module and return it.
|
81 |
+
"""
|
82 |
+
for p in module.parameters():
|
83 |
+
p.detach().zero_()
|
84 |
+
return module
|
85 |
+
|
86 |
+
|
87 |
+
def scale_module(module, scale):
|
88 |
+
"""
|
89 |
+
Scale the parameters of a module and return it.
|
90 |
+
"""
|
91 |
+
for p in module.parameters():
|
92 |
+
p.detach().mul_(scale)
|
93 |
+
return module
|
94 |
+
|
95 |
+
|
96 |
+
def mean_flat(tensor):
|
97 |
+
"""
|
98 |
+
Take the mean over all non-batch dimensions.
|
99 |
+
"""
|
100 |
+
return tensor.mean(dim=list(range(1, len(tensor.shape))))
|
101 |
+
|
102 |
+
|
103 |
+
def normalization(channels):
|
104 |
+
"""
|
105 |
+
Make a standard normalization layer.
|
106 |
+
|
107 |
+
:param channels: number of input channels.
|
108 |
+
:return: an nn.Module for normalization.
|
109 |
+
"""
|
110 |
+
return GroupNorm32(32, channels)
|
111 |
+
|
112 |
+
|
113 |
+
def timestep_embedding(timesteps, dim, max_period=10000):
|
114 |
+
"""
|
115 |
+
Create sinusoidal timestep embeddings.
|
116 |
+
|
117 |
+
:param timesteps: a 1-D Tensor of N indices, one per batch element.
|
118 |
+
These may be fractional.
|
119 |
+
:param dim: the dimension of the output.
|
120 |
+
:param max_period: controls the minimum frequency of the embeddings.
|
121 |
+
:return: an [N x dim] Tensor of positional embeddings.
|
122 |
+
"""
|
123 |
+
half = dim // 2
|
124 |
+
freqs = th.exp(
|
125 |
+
-math.log(max_period) * th.arange(start=0, end=half, dtype=th.float32) / half
|
126 |
+
).to(device=timesteps.device)
|
127 |
+
args = timesteps[:, None].float() * freqs[None]
|
128 |
+
embedding = th.cat([th.cos(args), th.sin(args)], dim=-1)
|
129 |
+
if dim % 2:
|
130 |
+
embedding = th.cat([embedding, th.zeros_like(embedding[:, :1])], dim=-1)
|
131 |
+
return embedding
|
132 |
+
|
133 |
+
|
134 |
+
def checkpoint(func, inputs, params, flag):
|
135 |
+
"""
|
136 |
+
Evaluate a function without caching intermediate activations, allowing for
|
137 |
+
reduced memory at the expense of extra compute in the backward pass.
|
138 |
+
|
139 |
+
:param func: the function to evaluate.
|
140 |
+
:param inputs: the argument sequence to pass to `func`.
|
141 |
+
:param params: a sequence of parameters `func` depends on but does not
|
142 |
+
explicitly take as arguments.
|
143 |
+
:param flag: if False, disable gradient checkpointing.
|
144 |
+
"""
|
145 |
+
if flag:
|
146 |
+
args = tuple(inputs) + tuple(params)
|
147 |
+
#return th.utils.checkpoint.checkpoint.apply(func, inputs)
|
148 |
+
return CheckpointFunction.apply(func, len(inputs), *args)
|
149 |
+
else:
|
150 |
+
return func(*inputs)
|
151 |
+
|
152 |
+
|
153 |
+
class CheckpointFunction(th.autograd.Function):
|
154 |
+
@staticmethod
|
155 |
+
def forward(ctx, run_function, length, *args):
|
156 |
+
ctx.run_function = run_function
|
157 |
+
ctx.input_tensors = list(args[:length])
|
158 |
+
ctx.input_params = list(args[length:])
|
159 |
+
breakpoint()
|
160 |
+
with th.no_grad():
|
161 |
+
output_tensors = ctx.run_function(*ctx.input_tensors)
|
162 |
+
return output_tensors
|
163 |
+
|
164 |
+
@staticmethod
|
165 |
+
def backward(ctx, *output_grads):
|
166 |
+
ctx.input_tensors = [x.detach().requires_grad_(True) for x in ctx.input_tensors]
|
167 |
+
with th.enable_grad():
|
168 |
+
# Fixes a bug where the first op in run_function modifies the
|
169 |
+
# Tensor storage in place, which is not allowed for detach()'d
|
170 |
+
# Tensors.
|
171 |
+
shallow_copies = [x.view_as(x) for x in ctx.input_tensors]
|
172 |
+
breakpoint()
|
173 |
+
output_tensors = ctx(*shallow_copies)
|
174 |
+
input_grads = th.autograd.grad(
|
175 |
+
output_tensors,
|
176 |
+
ctx.input_tensors + ctx.input_params,
|
177 |
+
output_grads,
|
178 |
+
allow_unused=True,
|
179 |
+
)
|
180 |
+
del ctx.input_tensors
|
181 |
+
del ctx.input_params
|
182 |
+
del output_tensors
|
183 |
+
return (None, None) + input_grads
|
diffusion_module/unet.py
ADDED
@@ -0,0 +1,1315 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from abc import abstractmethod
|
2 |
+
|
3 |
+
import math
|
4 |
+
|
5 |
+
import numpy as np
|
6 |
+
import torch as th
|
7 |
+
import torch.nn as nn
|
8 |
+
import torch.nn.functional as F
|
9 |
+
|
10 |
+
from .nn import (
|
11 |
+
SiLU,
|
12 |
+
checkpoint,
|
13 |
+
conv_nd,
|
14 |
+
linear,
|
15 |
+
avg_pool_nd,
|
16 |
+
zero_module,
|
17 |
+
normalization,
|
18 |
+
timestep_embedding,
|
19 |
+
convert_module_to_f16
|
20 |
+
)
|
21 |
+
|
22 |
+
from diffusers.configuration_utils import ConfigMixin, register_to_config
|
23 |
+
from diffusers.utils import BaseOutput
|
24 |
+
from diffusers.models.modeling_utils import ModelMixin
|
25 |
+
from dataclasses import dataclass
|
26 |
+
|
27 |
+
@dataclass
|
28 |
+
class UNet2DOutput(BaseOutput):
|
29 |
+
"""
|
30 |
+
Args:
|
31 |
+
sample (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)`):
|
32 |
+
Hidden states output. Output of last layer of model.
|
33 |
+
"""
|
34 |
+
|
35 |
+
sample: th.FloatTensor
|
36 |
+
|
37 |
+
|
38 |
+
class AttentionPool2d(nn.Module):
|
39 |
+
"""
|
40 |
+
Adapted from CLIP: https://github.com/openai/CLIP/blob/main/clip/model.py
|
41 |
+
"""
|
42 |
+
|
43 |
+
def __init__(
|
44 |
+
self,
|
45 |
+
spacial_dim: int,
|
46 |
+
embed_dim: int,
|
47 |
+
num_heads_channels: int,
|
48 |
+
output_dim: int = None,
|
49 |
+
):
|
50 |
+
super().__init__()
|
51 |
+
self.positional_embedding = nn.Parameter(
|
52 |
+
th.randn(embed_dim, spacial_dim ** 2 + 1) / embed_dim ** 0.5
|
53 |
+
)
|
54 |
+
self.qkv_proj = conv_nd(1, embed_dim, 3 * embed_dim, 1)
|
55 |
+
self.c_proj = conv_nd(1, embed_dim, output_dim or embed_dim, 1)
|
56 |
+
self.num_heads = embed_dim // num_heads_channels
|
57 |
+
self.attention = QKVAttention(self.num_heads)
|
58 |
+
|
59 |
+
def forward(self, x):
|
60 |
+
b, c, *_spatial = x.shape
|
61 |
+
x = x.reshape(b, c, -1) # NC(HW)
|
62 |
+
x = th.cat([x.mean(dim=-1, keepdim=True), x], dim=-1) # NC(HW+1)
|
63 |
+
x = x + self.positional_embedding[None, :, :].to(x.dtype) # NC(HW+1)
|
64 |
+
x = self.qkv_proj(x)
|
65 |
+
x = self.attention(x)
|
66 |
+
x = self.c_proj(x)
|
67 |
+
return x[:, :, 0]
|
68 |
+
|
69 |
+
|
70 |
+
class TimestepBlock(nn.Module):
|
71 |
+
"""
|
72 |
+
Any module where forward() takes timestep embeddings as a second argument.
|
73 |
+
"""
|
74 |
+
|
75 |
+
@abstractmethod
|
76 |
+
def forward(self, x, emb):
|
77 |
+
"""
|
78 |
+
Apply the module to `x` given `emb` timestep embeddings.
|
79 |
+
"""
|
80 |
+
|
81 |
+
class CondTimestepBlock(nn.Module):
|
82 |
+
"""
|
83 |
+
Any module where forward() takes timestep embeddings as a second argument.
|
84 |
+
"""
|
85 |
+
|
86 |
+
@abstractmethod
|
87 |
+
def forward(self, x, cond, emb):
|
88 |
+
"""
|
89 |
+
Apply the module to `x` given `emb` timestep embeddings.
|
90 |
+
"""
|
91 |
+
"""
|
92 |
+
class TimestepEmbedSequential(nn.Sequential, TimestepBlock, CondTimestepBlock):
|
93 |
+
|
94 |
+
def forward(self, x, cond, emb):
|
95 |
+
for layer in self:
|
96 |
+
if isinstance(layer, CondTimestepBlock):
|
97 |
+
x = layer(x, cond, emb)
|
98 |
+
elif isinstance(layer, TimestepBlock):
|
99 |
+
x = layer(x, emb)
|
100 |
+
else:
|
101 |
+
x = layer(x)
|
102 |
+
return x
|
103 |
+
"""
|
104 |
+
|
105 |
+
class TimestepEmbedSequential(nn.Sequential, TimestepBlock, CondTimestepBlock):
|
106 |
+
def forward(self, x, cond, emb):
|
107 |
+
outputs_list = [] # 创建一个空列表来存储第二个输出
|
108 |
+
for layer in self:
|
109 |
+
if isinstance(layer, CondTimestepBlock):
|
110 |
+
# 调用layer并检查输出是否为一个元组
|
111 |
+
result = layer(x, cond, emb)
|
112 |
+
if isinstance(result, tuple) and len(result) == 2:
|
113 |
+
x, additional_output = result
|
114 |
+
outputs_list.append(additional_output) # 将第二个输出添加到列表
|
115 |
+
else:
|
116 |
+
x = result
|
117 |
+
elif isinstance(layer, TimestepBlock):
|
118 |
+
x = layer(x, emb)
|
119 |
+
else:
|
120 |
+
x = layer(x)
|
121 |
+
|
122 |
+
if outputs_list == []:
|
123 |
+
return x
|
124 |
+
else:
|
125 |
+
return x, outputs_list # 返回最终的x和所有附加输出的列表
|
126 |
+
|
127 |
+
|
128 |
+
|
129 |
+
class Upsample(nn.Module):
|
130 |
+
"""
|
131 |
+
An upsampling layer with an optional convolution.
|
132 |
+
|
133 |
+
:param channels: channels in the inputs and outputs.
|
134 |
+
:param use_conv: a bool determining if a convolution is applied.
|
135 |
+
:param dims: determines if the signal is 1D, 2D, or 3D. If 3D, then
|
136 |
+
upsampling occurs in the inner-two dimensions.
|
137 |
+
"""
|
138 |
+
|
139 |
+
def __init__(self, channels, use_conv, dims=2, out_channels=None):
|
140 |
+
super().__init__()
|
141 |
+
self.channels = channels
|
142 |
+
self.out_channels = out_channels or channels
|
143 |
+
self.use_conv = use_conv
|
144 |
+
self.dims = dims
|
145 |
+
if use_conv:
|
146 |
+
self.conv = conv_nd(dims, self.channels, self.out_channels, 3, padding=1)
|
147 |
+
|
148 |
+
def forward(self, x):
|
149 |
+
assert x.shape[1] == self.channels
|
150 |
+
if self.dims == 3:
|
151 |
+
x = F.interpolate(
|
152 |
+
x, (x.shape[2], x.shape[3] * 2, x.shape[4] * 2), mode="nearest"
|
153 |
+
)
|
154 |
+
else:
|
155 |
+
x = F.interpolate(x, scale_factor=2, mode="nearest")
|
156 |
+
if self.use_conv:
|
157 |
+
x = self.conv(x)
|
158 |
+
return x
|
159 |
+
|
160 |
+
|
161 |
+
class Downsample(nn.Module):
|
162 |
+
"""
|
163 |
+
A downsampling layer with an optional convolution.
|
164 |
+
|
165 |
+
:param channels: channels in the inputs and outputs.
|
166 |
+
:param use_conv: a bool determining if a convolution is applied.
|
167 |
+
:param dims: determines if the signal is 1D, 2D, or 3D. If 3D, then
|
168 |
+
downsampling occurs in the inner-two dimensions.
|
169 |
+
"""
|
170 |
+
|
171 |
+
def __init__(self, channels, use_conv, dims=2, out_channels=None):
|
172 |
+
super().__init__()
|
173 |
+
self.channels = channels
|
174 |
+
self.out_channels = out_channels or channels
|
175 |
+
self.use_conv = use_conv
|
176 |
+
self.dims = dims
|
177 |
+
stride = 2 if dims != 3 else (1, 2, 2)
|
178 |
+
if use_conv:
|
179 |
+
self.op = conv_nd(
|
180 |
+
dims, self.channels, self.out_channels, 3, stride=stride, padding=1
|
181 |
+
)
|
182 |
+
else:
|
183 |
+
assert self.channels == self.out_channels
|
184 |
+
self.op = avg_pool_nd(dims, kernel_size=stride, stride=stride)
|
185 |
+
|
186 |
+
def forward(self, x):
|
187 |
+
assert x.shape[1] == self.channels
|
188 |
+
return self.op(x)
|
189 |
+
|
190 |
+
|
191 |
+
class SPADEGroupNorm(nn.Module):
|
192 |
+
def __init__(self, norm_nc, label_nc, eps = 1e-5,debug = False):
|
193 |
+
super().__init__()
|
194 |
+
self.debug = debug
|
195 |
+
self.norm = nn.GroupNorm(32, norm_nc, affine=False) # 32/16
|
196 |
+
|
197 |
+
self.eps = eps
|
198 |
+
nhidden = 128
|
199 |
+
self.mlp_shared = nn.Sequential(
|
200 |
+
nn.Conv2d(label_nc, nhidden, kernel_size=3, padding=1),
|
201 |
+
nn.ReLU(),
|
202 |
+
)
|
203 |
+
self.mlp_gamma = nn.Conv2d(nhidden, norm_nc, kernel_size=3, padding=1)
|
204 |
+
self.mlp_beta = nn.Conv2d(nhidden, norm_nc, kernel_size=3, padding=1)
|
205 |
+
|
206 |
+
def forward(self, x, segmap):
|
207 |
+
# Part 1. generate parameter-free normalized activations
|
208 |
+
x = self.norm(x)
|
209 |
+
|
210 |
+
# Part 2. produce scaling and bias conditioned on semantic map
|
211 |
+
segmap = F.interpolate(segmap, size=x.size()[2:], mode='nearest')
|
212 |
+
actv = self.mlp_shared(segmap)
|
213 |
+
gamma = self.mlp_gamma(actv)
|
214 |
+
beta = self.mlp_beta(actv)
|
215 |
+
|
216 |
+
# apply scale and bias
|
217 |
+
if self.debug:
|
218 |
+
return x * (1 + gamma) + beta, (beta.detach().cpu(), gamma.detach().cpu())
|
219 |
+
else:
|
220 |
+
return x * (1 + gamma) + beta
|
221 |
+
|
222 |
+
|
223 |
+
class AdaIN(nn.Module):
|
224 |
+
def __init__(self, num_features):
|
225 |
+
super().__init__()
|
226 |
+
self.instance_norm = th.nn.InstanceNorm2d(num_features, affine=False, track_running_stats=False)
|
227 |
+
|
228 |
+
def forward(self, x, alpha, gamma):
|
229 |
+
assert x.shape[:2] == alpha.shape[:2] == gamma.shape[:2]
|
230 |
+
norm = self.instance_norm(x)
|
231 |
+
return alpha * norm + gamma
|
232 |
+
|
233 |
+
class RESAILGroupNorm(nn.Module):
|
234 |
+
def __init__(self, norm_nc, label_nc, guidance_nc, eps = 1e-5):
|
235 |
+
super().__init__()
|
236 |
+
|
237 |
+
self.norm = nn.GroupNorm(32, norm_nc, affine=False) # 32/16
|
238 |
+
|
239 |
+
# SPADE
|
240 |
+
self.eps = eps
|
241 |
+
nhidden = 128
|
242 |
+
self.mask_mlp_shared = nn.Sequential(
|
243 |
+
nn.Conv2d(label_nc, nhidden, kernel_size=3, padding=1),
|
244 |
+
nn.ReLU(),
|
245 |
+
)
|
246 |
+
|
247 |
+
self.mask_mlp_gamma = nn.Conv2d(nhidden, norm_nc, kernel_size=3, padding=1)
|
248 |
+
self.mask_mlp_beta = nn.Conv2d(nhidden, norm_nc, kernel_size=3, padding=1)
|
249 |
+
|
250 |
+
|
251 |
+
# Guidance
|
252 |
+
|
253 |
+
self.conv_s = th.nn.Conv2d(label_nc, nhidden * 2, 3, 2)
|
254 |
+
self.pool_s = th.nn.AdaptiveAvgPool2d(1)
|
255 |
+
self.conv_s2 = th.nn.Conv2d(nhidden * 2, nhidden * 2, 1, 1)
|
256 |
+
|
257 |
+
self.conv1 = th.nn.Conv2d(guidance_nc, nhidden, 3, 1, padding=1)
|
258 |
+
self.adaIn1 = AdaIN(norm_nc * 2)
|
259 |
+
self.relu1 = nn.ReLU()
|
260 |
+
|
261 |
+
self.conv2 = th.nn.Conv2d(nhidden, nhidden, 3, 1, padding=1)
|
262 |
+
self.adaIn2 = AdaIN(norm_nc * 2)
|
263 |
+
self.relu2 = nn.ReLU()
|
264 |
+
self.conv3 = th.nn.Conv2d(nhidden, nhidden, 3, 1, padding=1)
|
265 |
+
|
266 |
+
self.guidance_mlp_gamma = nn.Conv2d(nhidden, norm_nc, kernel_size=3, padding=1)
|
267 |
+
self.guidance_mlp_beta = nn.Conv2d(nhidden, norm_nc, kernel_size=3, padding=1)
|
268 |
+
|
269 |
+
self.blending_gamma = nn.Parameter(th.zeros(1), requires_grad=True)
|
270 |
+
self.blending_beta = nn.Parameter(th.zeros(1), requires_grad=True)
|
271 |
+
self.norm_nc = norm_nc
|
272 |
+
|
273 |
+
def forward(self, x, segmap, guidance):
|
274 |
+
# Part 1. generate parameter-free normalized activations
|
275 |
+
x = self.norm(x)
|
276 |
+
# Part 2. produce scaling and bias conditioned on semantic map
|
277 |
+
segmap = F.interpolate(segmap, size=x.size()[2:], mode='nearest')
|
278 |
+
mask_actv = self.mask_mlp_shared(segmap)
|
279 |
+
mask_gamma = self.mask_mlp_gamma(mask_actv)
|
280 |
+
mask_beta = self.mask_mlp_beta(mask_actv)
|
281 |
+
|
282 |
+
|
283 |
+
# Part 3. produce scaling and bias conditioned on feature guidance
|
284 |
+
guidance = F.interpolate(guidance, size=x.size()[2:], mode='bilinear')
|
285 |
+
|
286 |
+
f_s_1 = self.conv_s(segmap)
|
287 |
+
c1 = self.pool_s(f_s_1)
|
288 |
+
c2 = self.conv_s2(c1)
|
289 |
+
|
290 |
+
f1 = self.conv1(guidance)
|
291 |
+
|
292 |
+
f1 = self.adaIn1(f1, c1[:, : 128, ...], c1[:, 128:, ...])
|
293 |
+
f2 = self.relu1(f1)
|
294 |
+
|
295 |
+
f2 = self.conv2(f2)
|
296 |
+
f2 = self.adaIn2(f2, c2[:, : 128, ...], c2[:, 128:, ...])
|
297 |
+
f2 = self.relu2(f2)
|
298 |
+
guidance_actv = self.conv3(f2)
|
299 |
+
|
300 |
+
guidance_gamma = self.guidance_mlp_gamma(guidance_actv)
|
301 |
+
guidance_beta = self.guidance_mlp_beta(guidance_actv)
|
302 |
+
|
303 |
+
gamma_alpha = F.sigmoid(self.blending_gamma)
|
304 |
+
beta_alpha = F.sigmoid(self.blending_beta)
|
305 |
+
|
306 |
+
gamma_final = gamma_alpha * guidance_gamma + (1 - gamma_alpha) * mask_gamma
|
307 |
+
beta_final = beta_alpha * guidance_beta + (1 - beta_alpha) * mask_beta
|
308 |
+
out = x * (1 + gamma_final) + beta_final
|
309 |
+
|
310 |
+
# apply scale and bias
|
311 |
+
return out
|
312 |
+
|
313 |
+
class SPMGroupNorm(nn.Module):
|
314 |
+
def __init__(self, norm_nc, label_nc, feature_nc, eps = 1e-5):
|
315 |
+
super().__init__()
|
316 |
+
print("use SPM")
|
317 |
+
|
318 |
+
self.norm = nn.GroupNorm(32, norm_nc, affine=False) # 32/16
|
319 |
+
|
320 |
+
# SPADE
|
321 |
+
self.eps = eps
|
322 |
+
nhidden = 128
|
323 |
+
self.mask_mlp_shared = nn.Sequential(
|
324 |
+
nn.Conv2d(label_nc, nhidden, kernel_size=3, padding=1),
|
325 |
+
nn.ReLU(),
|
326 |
+
)
|
327 |
+
|
328 |
+
self.mask_mlp_gamma1 = nn.Conv2d(nhidden, norm_nc, kernel_size=3, padding=1)
|
329 |
+
self.mask_mlp_beta1 = nn.Conv2d(nhidden, norm_nc, kernel_size=3, padding=1)
|
330 |
+
|
331 |
+
self.mask_mlp_gamma2 = nn.Conv2d(nhidden, norm_nc, kernel_size=3, padding=1)
|
332 |
+
self.mask_mlp_beta2 = nn.Conv2d(nhidden, norm_nc, kernel_size=3, padding=1)
|
333 |
+
|
334 |
+
|
335 |
+
# Feature
|
336 |
+
self.feature_mlp_shared = nn.Sequential(
|
337 |
+
nn.Conv2d(feature_nc, nhidden, kernel_size=3, padding=1),
|
338 |
+
nn.ReLU(),
|
339 |
+
)
|
340 |
+
|
341 |
+
self.feature_mlp_gamma1 = nn.Conv2d(nhidden, norm_nc, kernel_size=3, padding=1)
|
342 |
+
self.feature_mlp_beta1 = nn.Conv2d(nhidden, norm_nc, kernel_size=3, padding=1)
|
343 |
+
|
344 |
+
|
345 |
+
def forward(self, x, segmap, guidance):
|
346 |
+
# Part 1. generate parameter-free normalized activations
|
347 |
+
x = self.norm(x)
|
348 |
+
# Part 2. produce scaling and bias conditioned on semantic map
|
349 |
+
segmap = F.interpolate(segmap, size=x.size()[2:], mode='nearest')
|
350 |
+
mask_actv = self.mask_mlp_shared(segmap)
|
351 |
+
mask_gamma1 = self.mask_mlp_gamma1(mask_actv)
|
352 |
+
mask_beta1 = self.mask_mlp_beta1(mask_actv)
|
353 |
+
|
354 |
+
mask_gamma2 = self.mask_mlp_gamma2(mask_actv)
|
355 |
+
mask_beta2 = self.mask_mlp_beta2(mask_actv)
|
356 |
+
|
357 |
+
|
358 |
+
# Part 3. produce scaling and bias conditioned on feature guidance
|
359 |
+
guidance = F.interpolate(guidance, size=x.size()[2:], mode='bilinear')
|
360 |
+
feature_actv = self.feature_mlp_shared(guidance)
|
361 |
+
feature_gamma1 = self.feature_mlp_gamma1(feature_actv)
|
362 |
+
feature_beta1 = self.feature_mlp_beta1(feature_actv)
|
363 |
+
|
364 |
+
gamma_final = feature_gamma1 * (1 + mask_gamma1) + mask_beta1
|
365 |
+
beta_final = feature_beta1 * (1 + mask_gamma2) + mask_beta2
|
366 |
+
|
367 |
+
out = x * (1 + gamma_final) + beta_final
|
368 |
+
|
369 |
+
# apply scale and bias
|
370 |
+
return out
|
371 |
+
|
372 |
+
|
373 |
+
class ResBlock(TimestepBlock):
|
374 |
+
"""
|
375 |
+
A residual block that can optionally change the number of channels.
|
376 |
+
|
377 |
+
:param channels: the number of input channels.
|
378 |
+
:param emb_channels: the number of timestep embedding channels.
|
379 |
+
:param dropout: the rate of dropout.
|
380 |
+
:param out_channels: if specified, the number of out channels.
|
381 |
+
:param use_conv: if True and out_channels is specified, use a spatial
|
382 |
+
convolution instead of a smaller 1x1 convolution to change the
|
383 |
+
channels in the skip connection.
|
384 |
+
:param dims: determines if the signal is 1D, 2D, or 3D.
|
385 |
+
:param use_checkpoint: if True, use gradient checkpointing on this module.
|
386 |
+
:param up: if True, use this block for upsampling.
|
387 |
+
:param down: if True, use this block for downsampling.
|
388 |
+
"""
|
389 |
+
|
390 |
+
def __init__(
|
391 |
+
self,
|
392 |
+
channels,
|
393 |
+
emb_channels,
|
394 |
+
dropout,
|
395 |
+
out_channels=None,
|
396 |
+
use_conv=False,
|
397 |
+
use_scale_shift_norm=False,
|
398 |
+
dims=2,
|
399 |
+
use_checkpoint=False,
|
400 |
+
up=False,
|
401 |
+
down=False,
|
402 |
+
):
|
403 |
+
super().__init__()
|
404 |
+
self.channels = channels
|
405 |
+
self.emb_channels = emb_channels
|
406 |
+
self.dropout = dropout
|
407 |
+
self.out_channels = out_channels or channels
|
408 |
+
self.use_conv = use_conv
|
409 |
+
self.use_checkpoint = use_checkpoint
|
410 |
+
self.use_scale_shift_norm = use_scale_shift_norm
|
411 |
+
|
412 |
+
self.in_layers = nn.Sequential(
|
413 |
+
normalization(channels),
|
414 |
+
SiLU(),
|
415 |
+
conv_nd(dims, channels, self.out_channels, 3, padding=1),
|
416 |
+
)
|
417 |
+
|
418 |
+
self.updown = up or down
|
419 |
+
|
420 |
+
if up:
|
421 |
+
self.h_upd = Upsample(channels, False, dims)
|
422 |
+
self.x_upd = Upsample(channels, False, dims)
|
423 |
+
elif down:
|
424 |
+
self.h_upd = Downsample(channels, False, dims)
|
425 |
+
self.x_upd = Downsample(channels, False, dims)
|
426 |
+
else:
|
427 |
+
self.h_upd = self.x_upd = nn.Identity()
|
428 |
+
|
429 |
+
self.emb_layers = nn.Sequential(
|
430 |
+
SiLU(),
|
431 |
+
linear(
|
432 |
+
emb_channels,
|
433 |
+
2 * self.out_channels if use_scale_shift_norm else self.out_channels,
|
434 |
+
),
|
435 |
+
)
|
436 |
+
self.out_layers = nn.Sequential(
|
437 |
+
normalization(self.out_channels),
|
438 |
+
SiLU(),
|
439 |
+
nn.Dropout(p=dropout),
|
440 |
+
zero_module(
|
441 |
+
conv_nd(dims, self.out_channels, self.out_channels, 3, padding=1)
|
442 |
+
),
|
443 |
+
)
|
444 |
+
|
445 |
+
if self.out_channels == channels:
|
446 |
+
self.skip_connection = nn.Identity()
|
447 |
+
elif use_conv:
|
448 |
+
self.skip_connection = conv_nd(
|
449 |
+
dims, channels, self.out_channels, 3, padding=1
|
450 |
+
)
|
451 |
+
else:
|
452 |
+
self.skip_connection = conv_nd(dims, channels, self.out_channels, 1)
|
453 |
+
|
454 |
+
def forward(self, x, emb):
|
455 |
+
"""
|
456 |
+
Apply the block to a Tensor, conditioned on a timestep embedding.
|
457 |
+
|
458 |
+
:param x: an [N x C x ...] Tensor of features.
|
459 |
+
:param emb: an [N x emb_channels] Tensor of timestep embeddings.
|
460 |
+
:return: an [N x C x ...] Tensor of outputs.
|
461 |
+
"""
|
462 |
+
|
463 |
+
return th.utils.checkpoint.checkpoint(self._forward, x ,emb)
|
464 |
+
# return checkpoint(
|
465 |
+
# self._forward, (x, emb), self.parameters(), self.use_checkpoint
|
466 |
+
# )
|
467 |
+
|
468 |
+
def _forward(self, x, emb):
|
469 |
+
if self.updown:
|
470 |
+
in_rest, in_conv = self.in_layers[:-1], self.in_layers[-1]
|
471 |
+
h = in_rest(x)
|
472 |
+
h = self.h_upd(h)
|
473 |
+
x = self.x_upd(x)
|
474 |
+
h = in_conv(h)
|
475 |
+
else:
|
476 |
+
h = self.in_layers(x)
|
477 |
+
emb_out = self.emb_layers(emb)#.type(h.dtype)
|
478 |
+
while len(emb_out.shape) < len(h.shape):
|
479 |
+
emb_out = emb_out[..., None]
|
480 |
+
if self.use_scale_shift_norm:
|
481 |
+
out_norm, out_rest = self.out_layers[0], self.out_layers[1:]
|
482 |
+
scale, shift = th.chunk(emb_out, 2, dim=1)
|
483 |
+
h = out_norm(h) * (1 + scale) + shift
|
484 |
+
h = out_rest(h)
|
485 |
+
else:
|
486 |
+
h = h + emb_out
|
487 |
+
h = self.out_layers(h)
|
488 |
+
return self.skip_connection(x) + h
|
489 |
+
|
490 |
+
class SDMResBlock(CondTimestepBlock):
|
491 |
+
"""
|
492 |
+
A residual block that can optionally change the number of channels.
|
493 |
+
|
494 |
+
:param channels: the number of input channels.
|
495 |
+
:param emb_channels: the number of timestep embedding channels.
|
496 |
+
:param dropout: the rate of dropout.
|
497 |
+
:param out_channels: if specified, the number of out channels.
|
498 |
+
:param use_conv: if True and out_channels is specified, use a spatial
|
499 |
+
convolution instead of a smaller 1x1 convolution to change the
|
500 |
+
channels in the skip connection.
|
501 |
+
:param dims: determines if the signal is 1D, 2D, or 3D.
|
502 |
+
:param use_checkpoint: if True, use gradient checkpointing on this module.
|
503 |
+
:param up: if True, use this block for upsampling.
|
504 |
+
:param down: if True, use this block for downsampling.
|
505 |
+
"""
|
506 |
+
|
507 |
+
def __init__(
|
508 |
+
self,
|
509 |
+
channels,
|
510 |
+
emb_channels,
|
511 |
+
dropout,
|
512 |
+
c_channels=3,
|
513 |
+
out_channels=None,
|
514 |
+
use_conv=False,
|
515 |
+
use_scale_shift_norm=False,
|
516 |
+
dims=2,
|
517 |
+
use_checkpoint=False,
|
518 |
+
up=False,
|
519 |
+
down=False,
|
520 |
+
SPADE_type = "spade",
|
521 |
+
guidance_nc = None,
|
522 |
+
debug = False
|
523 |
+
):
|
524 |
+
super().__init__()
|
525 |
+
self.channels = channels
|
526 |
+
self.guidance_nc = guidance_nc
|
527 |
+
self.emb_channels = emb_channels
|
528 |
+
self.dropout = dropout
|
529 |
+
self.out_channels = out_channels or channels
|
530 |
+
self.use_conv = use_conv
|
531 |
+
self.use_checkpoint = use_checkpoint
|
532 |
+
self.use_scale_shift_norm = use_scale_shift_norm
|
533 |
+
self.SPADE_type = SPADE_type
|
534 |
+
self.debug = debug
|
535 |
+
if self.SPADE_type == "spade":
|
536 |
+
self.in_norm = SPADEGroupNorm(channels, c_channels, debug=self.debug)
|
537 |
+
elif self.SPADE_type == "RESAIL":
|
538 |
+
self.in_norm = RESAILGroupNorm(channels, c_channels, guidance_nc)
|
539 |
+
elif self.SPADE_type == "SPM":
|
540 |
+
self.in_norm = SPMGroupNorm(channels, c_channels, guidance_nc)
|
541 |
+
self.in_layers = nn.Sequential(
|
542 |
+
SiLU(),
|
543 |
+
conv_nd(dims, channels, self.out_channels, 3, padding=1),
|
544 |
+
)
|
545 |
+
|
546 |
+
self.updown = up or down
|
547 |
+
|
548 |
+
if up:
|
549 |
+
self.h_upd = Upsample(channels, False, dims)
|
550 |
+
self.x_upd = Upsample(channels, False, dims)
|
551 |
+
elif down:
|
552 |
+
self.h_upd = Downsample(channels, False, dims)
|
553 |
+
self.x_upd = Downsample(channels, False, dims)
|
554 |
+
else:
|
555 |
+
self.h_upd = self.x_upd = nn.Identity()
|
556 |
+
|
557 |
+
self.emb_layers = nn.Sequential(
|
558 |
+
SiLU(),
|
559 |
+
linear(
|
560 |
+
emb_channels,
|
561 |
+
2 * self.out_channels if use_scale_shift_norm else self.out_channels,
|
562 |
+
),
|
563 |
+
)
|
564 |
+
|
565 |
+
if self.SPADE_type == "spade":
|
566 |
+
self.out_norm = SPADEGroupNorm(self.out_channels, c_channels,debug=self.debug)
|
567 |
+
elif self.SPADE_type == "RESAIL":
|
568 |
+
self.out_norm = RESAILGroupNorm(self.out_channels, c_channels, guidance_nc)
|
569 |
+
elif self.SPADE_type == "SPM":
|
570 |
+
self.out_norm = SPMGroupNorm(self.out_channels, c_channels, guidance_nc)
|
571 |
+
|
572 |
+
self.out_layers = nn.Sequential(
|
573 |
+
SiLU(),
|
574 |
+
nn.Dropout(p=dropout),
|
575 |
+
zero_module(
|
576 |
+
conv_nd(dims, self.out_channels, self.out_channels, 3, padding=1)
|
577 |
+
),
|
578 |
+
)
|
579 |
+
|
580 |
+
if self.out_channels == channels:
|
581 |
+
self.skip_connection = nn.Identity()
|
582 |
+
elif use_conv:
|
583 |
+
self.skip_connection = conv_nd(
|
584 |
+
dims, channels, self.out_channels, 3, padding=1
|
585 |
+
)
|
586 |
+
else:
|
587 |
+
self.skip_connection = conv_nd(dims, channels, self.out_channels, 1)
|
588 |
+
|
589 |
+
def forward(self, x, cond, emb):
|
590 |
+
"""
|
591 |
+
Apply the block to a Tensor, conditioned on a timestep embedding.
|
592 |
+
|
593 |
+
:param x: an [N x C x ...] Tensor of features.
|
594 |
+
:param emb: an [N x emb_channels] Tensor of timestep embeddings.
|
595 |
+
:return: an [N x C x ...] Tensor of outputs.
|
596 |
+
"""
|
597 |
+
return th.utils.checkpoint.checkpoint(self._forward, x, cond, emb)
|
598 |
+
# return checkpoint(
|
599 |
+
# self._forward, (x, cond, emb), self.parameters(), self.use_checkpoint
|
600 |
+
# )
|
601 |
+
|
602 |
+
def _forward(self, x, cond, emb):
|
603 |
+
if self.SPADE_type == "RESAIL" or self.SPADE_type == "SPM":
|
604 |
+
assert self.guidance_nc is not None, "Please set guidance_nc when you use RESAIL"
|
605 |
+
guidance = x[: ,x.shape[1] - self.guidance_nc:, ...]
|
606 |
+
else:
|
607 |
+
guidance = None
|
608 |
+
if self.updown:
|
609 |
+
in_rest, in_conv = self.in_layers[:-1], self.in_layers[-1]
|
610 |
+
if self.SPADE_type == "spade":
|
611 |
+
if not self.debug:
|
612 |
+
h = self.in_norm(x, cond)
|
613 |
+
else:
|
614 |
+
h, (b1,g1) = self.in_norm(x, cond)
|
615 |
+
elif self.SPADE_type == "RESAIL" or self.SPADE_type == "SPM":
|
616 |
+
h = self.in_norm(x, cond, guidance)
|
617 |
+
|
618 |
+
h = in_rest(h)
|
619 |
+
h = self.h_upd(h)
|
620 |
+
x = self.x_upd(x)
|
621 |
+
h = in_conv(h)
|
622 |
+
else:
|
623 |
+
if self.SPADE_type == "spade":
|
624 |
+
if not self.debug:
|
625 |
+
h = self.in_norm(x, cond)
|
626 |
+
else:
|
627 |
+
h, (b1,g1) = self.in_norm(x, cond)
|
628 |
+
elif self.SPADE_type == "RESAIL" or self.SPADE_type == "SPM":
|
629 |
+
h = self.in_norm(x, cond, guidance)
|
630 |
+
h = self.in_layers(h)
|
631 |
+
|
632 |
+
emb_out = self.emb_layers(emb)#.type(h.dtype)
|
633 |
+
while len(emb_out.shape) < len(h.shape):
|
634 |
+
emb_out = emb_out[..., None]
|
635 |
+
if self.use_scale_shift_norm:
|
636 |
+
scale, shift = th.chunk(emb_out, 2, dim=1)
|
637 |
+
if self.SPADE_type == "spade":
|
638 |
+
if not self.debug:
|
639 |
+
h = self.out_norm(h, cond)
|
640 |
+
else:
|
641 |
+
h, (b2,g2) = self.out_norm(h, cond)
|
642 |
+
elif self.SPADE_type == "RESAIL" or self.SPADE_type == "SPM":
|
643 |
+
h = self.out_norm(h, cond, guidance)
|
644 |
+
|
645 |
+
h = h * (1 + scale) + shift
|
646 |
+
h = self.out_layers(h)
|
647 |
+
else:
|
648 |
+
h = h + emb_out
|
649 |
+
if self.SPADE_type == "spade":
|
650 |
+
h = self.out_norm(h, cond)
|
651 |
+
elif self.SPADE_type == "RESAIL" or self.SPADE_type == "SPM":
|
652 |
+
h = self.out_norm(x, cond, guidance)
|
653 |
+
|
654 |
+
h = self.out_layers(h)
|
655 |
+
if self.debug:
|
656 |
+
extra = {(b1,g1),(b2,g2)}
|
657 |
+
return self.skip_connection(x) + h, extra
|
658 |
+
else:
|
659 |
+
return self.skip_connection(x) + h
|
660 |
+
|
661 |
+
class AttentionBlock(nn.Module):
|
662 |
+
"""
|
663 |
+
An attention block that allows spatial positions to attend to each other.
|
664 |
+
|
665 |
+
Originally ported from here, but adapted to the N-d case.
|
666 |
+
https://github.com/hojonathanho/diffusion/blob/1e0dceb3b3495bbe19116a5e1b3596cd0706c543/diffusion_tf/models/unet.py#L66.
|
667 |
+
"""
|
668 |
+
|
669 |
+
def __init__(
|
670 |
+
self,
|
671 |
+
channels,
|
672 |
+
num_heads=1,
|
673 |
+
num_head_channels=-1,
|
674 |
+
use_checkpoint=False,
|
675 |
+
use_new_attention_order=False,
|
676 |
+
):
|
677 |
+
super().__init__()
|
678 |
+
self.channels = channels
|
679 |
+
if num_head_channels == -1:
|
680 |
+
self.num_heads = num_heads
|
681 |
+
else:
|
682 |
+
assert (
|
683 |
+
channels % num_head_channels == 0
|
684 |
+
), f"q,k,v channels {channels} is not divisible by num_head_channels {num_head_channels}"
|
685 |
+
self.num_heads = channels // num_head_channels
|
686 |
+
self.use_checkpoint = use_checkpoint
|
687 |
+
self.norm = normalization(channels)
|
688 |
+
self.qkv = conv_nd(1, channels, channels * 3, 1)
|
689 |
+
if use_new_attention_order:
|
690 |
+
# split qkv before split heads
|
691 |
+
self.attention = QKVAttention(self.num_heads)
|
692 |
+
else:
|
693 |
+
# split heads before split qkv
|
694 |
+
self.attention = QKVAttentionLegacy(self.num_heads)
|
695 |
+
|
696 |
+
self.proj_out = zero_module(conv_nd(1, channels, channels, 1))
|
697 |
+
|
698 |
+
def forward(self, x):
|
699 |
+
return th.utils.checkpoint.checkpoint(self._forward, x)
|
700 |
+
#return checkpoint(self._forward, (x,), self.parameters(), self.use_checkpoint)
|
701 |
+
|
702 |
+
def _forward(self, x):
|
703 |
+
b, c, *spatial = x.shape
|
704 |
+
x = x.reshape(b, c, -1)
|
705 |
+
qkv = self.qkv(self.norm(x))
|
706 |
+
h = self.attention(qkv)
|
707 |
+
h = self.proj_out(h)
|
708 |
+
return (x + h).reshape(b, c, *spatial)
|
709 |
+
|
710 |
+
|
711 |
+
def count_flops_attn(model, _x, y):
|
712 |
+
"""
|
713 |
+
A counter for the `thop` package to count the operations in an
|
714 |
+
attention operation.
|
715 |
+
Meant to be used like:
|
716 |
+
macs, params = thop.profile(
|
717 |
+
model,
|
718 |
+
inputs=(inputs, timestamps),
|
719 |
+
custom_ops={QKVAttention: QKVAttention.count_flops},
|
720 |
+
)
|
721 |
+
"""
|
722 |
+
b, c, *spatial = y[0].shape
|
723 |
+
num_spatial = int(np.prod(spatial))
|
724 |
+
# We perform two matmuls with the same number of ops.
|
725 |
+
# The first computes the weight matrix, the second computes
|
726 |
+
# the combination of the value vectors.
|
727 |
+
matmul_ops = 2 * b * (num_spatial ** 2) * c
|
728 |
+
model.total_ops += th.DoubleTensor([matmul_ops])
|
729 |
+
|
730 |
+
|
731 |
+
class QKVAttentionLegacy(nn.Module):
|
732 |
+
"""
|
733 |
+
A module which performs QKV attention. Matches legacy QKVAttention + input/ouput heads shaping
|
734 |
+
"""
|
735 |
+
|
736 |
+
def __init__(self, n_heads):
|
737 |
+
super().__init__()
|
738 |
+
self.n_heads = n_heads
|
739 |
+
|
740 |
+
def forward(self, qkv):
|
741 |
+
"""
|
742 |
+
Apply QKV attention.
|
743 |
+
|
744 |
+
:param qkv: an [N x (H * 3 * C) x T] tensor of Qs, Ks, and Vs.
|
745 |
+
:return: an [N x (H * C) x T] tensor after attention.
|
746 |
+
"""
|
747 |
+
bs, width, length = qkv.shape
|
748 |
+
assert width % (3 * self.n_heads) == 0
|
749 |
+
ch = width // (3 * self.n_heads)
|
750 |
+
q, k, v = qkv.reshape(bs * self.n_heads, ch * 3, length).split(ch, dim=1)
|
751 |
+
scale = 1 / math.sqrt(math.sqrt(ch))
|
752 |
+
weight = th.einsum(
|
753 |
+
"bct,bcs->bts", q * scale, k * scale
|
754 |
+
) # More stable with f16 than dividing afterwards
|
755 |
+
weight = th.softmax(weight.float(), dim=-1).type(weight.dtype)
|
756 |
+
a = th.einsum("bts,bcs->bct", weight, v)
|
757 |
+
return a.reshape(bs, -1, length)
|
758 |
+
|
759 |
+
@staticmethod
|
760 |
+
def count_flops(model, _x, y):
|
761 |
+
return count_flops_attn(model, _x, y)
|
762 |
+
|
763 |
+
|
764 |
+
class QKVAttention(nn.Module):
|
765 |
+
"""
|
766 |
+
A module which performs QKV attention and splits in a different order.
|
767 |
+
"""
|
768 |
+
|
769 |
+
def __init__(self, n_heads):
|
770 |
+
super().__init__()
|
771 |
+
self.n_heads = n_heads
|
772 |
+
|
773 |
+
def forward(self, qkv):
|
774 |
+
"""
|
775 |
+
Apply QKV attention.
|
776 |
+
|
777 |
+
:param qkv: an [N x (3 * H * C) x T] tensor of Qs, Ks, and Vs.
|
778 |
+
:return: an [N x (H * C) x T] tensor after attention.
|
779 |
+
"""
|
780 |
+
bs, width, length = qkv.shape
|
781 |
+
assert width % (3 * self.n_heads) == 0
|
782 |
+
ch = width // (3 * self.n_heads)
|
783 |
+
q, k, v = qkv.chunk(3, dim=1)
|
784 |
+
scale = 1 / math.sqrt(math.sqrt(ch))
|
785 |
+
weight = th.einsum(
|
786 |
+
"bct,bcs->bts",
|
787 |
+
(q * scale).view(bs * self.n_heads, ch, length),
|
788 |
+
(k * scale).view(bs * self.n_heads, ch, length),
|
789 |
+
) # More stable with f16 than dividing afterwards
|
790 |
+
weight = th.softmax(weight.float(), dim=-1).type(weight.dtype)
|
791 |
+
a = th.einsum("bts,bcs->bct", weight, v.reshape(bs * self.n_heads, ch, length))
|
792 |
+
return a.reshape(bs, -1, length)
|
793 |
+
|
794 |
+
@staticmethod
|
795 |
+
def count_flops(model, _x, y):
|
796 |
+
return count_flops_attn(model, _x, y)
|
797 |
+
|
798 |
+
|
799 |
+
class UNetModel(ModelMixin, ConfigMixin):
|
800 |
+
"""
|
801 |
+
The full UNet model with attention and timestep embedding.
|
802 |
+
|
803 |
+
:param in_channels: channels in the input Tensor.
|
804 |
+
:param model_channels: base channel count for the model.
|
805 |
+
:param out_channels: channels in the output Tensor.
|
806 |
+
:param num_res_blocks: number of residual blocks per downsample.
|
807 |
+
:param attention_resolutions: a collection of downsample rates at which
|
808 |
+
attention will take place. May be a set, list, or tuple.
|
809 |
+
For example, if this contains 4, then at 4x downsampling, attention
|
810 |
+
will be used.
|
811 |
+
:param dropout: the dropout probability.
|
812 |
+
:param channel_mult: channel multiplier for each level of the UNet.
|
813 |
+
:param conv_resample: if True, use learned convolutions for upsampling and
|
814 |
+
downsampling.
|
815 |
+
:param dims: determines if the signal is 1D, 2D, or 3D.
|
816 |
+
:param num_classes: if specified (as an int), then this model will be
|
817 |
+
class-conditional with `num_classes` classes.
|
818 |
+
:param use_checkpoint: use gradient checkpointing to reduce memory usage.
|
819 |
+
:param num_heads: the number of attention heads in each attention layer.
|
820 |
+
:param num_heads_channels: if specified, ignore num_heads and instead use
|
821 |
+
a fixed channel width per attention head.
|
822 |
+
:param num_heads_upsample: works with num_heads to set a different number
|
823 |
+
of heads for upsampling. Deprecated.
|
824 |
+
:param use_scale_shift_norm: use a FiLM-like conditioning mechanism.
|
825 |
+
:param resblock_updown: use residual blocks for up/downsampling.
|
826 |
+
:param use_new_attention_order: use a different attention pattern for potentially
|
827 |
+
increased efficiency.
|
828 |
+
"""
|
829 |
+
|
830 |
+
_supports_gradient_checkpointing = True
|
831 |
+
@register_to_config
|
832 |
+
def __init__(
|
833 |
+
self,
|
834 |
+
image_size,
|
835 |
+
in_channels,
|
836 |
+
model_channels,
|
837 |
+
out_channels,
|
838 |
+
num_res_blocks,
|
839 |
+
attention_resolutions,
|
840 |
+
dropout=0,
|
841 |
+
channel_mult=(1, 2, 4, 8),
|
842 |
+
conv_resample=True,
|
843 |
+
dims=2,
|
844 |
+
num_classes=None,
|
845 |
+
use_checkpoint=False,
|
846 |
+
use_fp16=True,
|
847 |
+
num_heads=1,
|
848 |
+
num_head_channels=-1,
|
849 |
+
num_heads_upsample=-1,
|
850 |
+
use_scale_shift_norm=False,
|
851 |
+
resblock_updown=False,
|
852 |
+
use_new_attention_order=False,
|
853 |
+
mask_emb="resize",
|
854 |
+
SPADE_type="spade",
|
855 |
+
debug = False
|
856 |
+
):
|
857 |
+
super().__init__()
|
858 |
+
|
859 |
+
if num_heads_upsample == -1:
|
860 |
+
num_heads_upsample = num_heads
|
861 |
+
|
862 |
+
self.sample_size = image_size
|
863 |
+
self.in_channels = in_channels
|
864 |
+
self.model_channels = model_channels
|
865 |
+
self.out_channels = out_channels
|
866 |
+
self.num_res_blocks = num_res_blocks
|
867 |
+
self.attention_resolutions = attention_resolutions
|
868 |
+
self.dropout = dropout
|
869 |
+
self.channel_mult = channel_mult
|
870 |
+
self.conv_resample = conv_resample
|
871 |
+
self.num_classes = num_classes
|
872 |
+
self.use_checkpoint = use_checkpoint
|
873 |
+
self.num_heads = num_heads
|
874 |
+
self.num_head_channels = num_head_channels
|
875 |
+
self.num_heads_upsample = num_heads_upsample
|
876 |
+
|
877 |
+
self.debug = debug
|
878 |
+
|
879 |
+
self.mask_emb = mask_emb
|
880 |
+
|
881 |
+
time_embed_dim = model_channels * 4
|
882 |
+
self.time_embed = nn.Sequential(
|
883 |
+
linear(model_channels, time_embed_dim),
|
884 |
+
SiLU(),
|
885 |
+
linear(time_embed_dim, time_embed_dim),
|
886 |
+
)
|
887 |
+
|
888 |
+
ch = input_ch = int(channel_mult[0] * model_channels)
|
889 |
+
self.input_blocks = nn.ModuleList(
|
890 |
+
[TimestepEmbedSequential(conv_nd(dims, in_channels, ch, 3, padding=1))] #ch=256
|
891 |
+
)
|
892 |
+
self._feature_size = ch
|
893 |
+
input_block_chans = [ch]
|
894 |
+
ds = 1
|
895 |
+
for level, mult in enumerate(channel_mult):
|
896 |
+
for _ in range(num_res_blocks):
|
897 |
+
layers = [
|
898 |
+
ResBlock(
|
899 |
+
ch,
|
900 |
+
time_embed_dim,
|
901 |
+
dropout,
|
902 |
+
out_channels=int(mult * model_channels),
|
903 |
+
dims=dims,
|
904 |
+
use_checkpoint=use_checkpoint,
|
905 |
+
use_scale_shift_norm=use_scale_shift_norm,
|
906 |
+
)
|
907 |
+
]
|
908 |
+
ch = int(mult * model_channels)
|
909 |
+
#print(ds)
|
910 |
+
if ds in attention_resolutions:
|
911 |
+
layers.append(
|
912 |
+
AttentionBlock(
|
913 |
+
ch,
|
914 |
+
use_checkpoint=use_checkpoint,
|
915 |
+
num_heads=num_heads,
|
916 |
+
num_head_channels=num_head_channels,
|
917 |
+
use_new_attention_order=use_new_attention_order,
|
918 |
+
)
|
919 |
+
)
|
920 |
+
self.input_blocks.append(TimestepEmbedSequential(*layers))
|
921 |
+
self._feature_size += ch
|
922 |
+
input_block_chans.append(ch)
|
923 |
+
if level != len(channel_mult) - 1:
|
924 |
+
out_ch = ch
|
925 |
+
self.input_blocks.append(
|
926 |
+
TimestepEmbedSequential(
|
927 |
+
ResBlock(
|
928 |
+
ch,
|
929 |
+
time_embed_dim,
|
930 |
+
dropout,
|
931 |
+
out_channels=out_ch,
|
932 |
+
dims=dims,
|
933 |
+
use_checkpoint=use_checkpoint,
|
934 |
+
use_scale_shift_norm=use_scale_shift_norm,
|
935 |
+
down=True,
|
936 |
+
)
|
937 |
+
if resblock_updown
|
938 |
+
else Downsample(
|
939 |
+
ch, conv_resample, dims=dims, out_channels=out_ch
|
940 |
+
)
|
941 |
+
)
|
942 |
+
)
|
943 |
+
ch = out_ch
|
944 |
+
input_block_chans.append(ch)
|
945 |
+
ds *= 2
|
946 |
+
self._feature_size += ch
|
947 |
+
self.middle_block = TimestepEmbedSequential(
|
948 |
+
SDMResBlock(
|
949 |
+
ch,
|
950 |
+
time_embed_dim,
|
951 |
+
dropout,
|
952 |
+
c_channels=num_classes if mask_emb == "resize" else num_classes*4,
|
953 |
+
dims=dims,
|
954 |
+
use_checkpoint=use_checkpoint,
|
955 |
+
use_scale_shift_norm=use_scale_shift_norm,
|
956 |
+
),
|
957 |
+
AttentionBlock(
|
958 |
+
ch,
|
959 |
+
use_checkpoint=use_checkpoint,
|
960 |
+
num_heads=num_heads,
|
961 |
+
num_head_channels=num_head_channels,
|
962 |
+
use_new_attention_order=use_new_attention_order,
|
963 |
+
),
|
964 |
+
SDMResBlock(
|
965 |
+
ch,
|
966 |
+
time_embed_dim,
|
967 |
+
dropout,
|
968 |
+
c_channels=num_classes if mask_emb == "resize" else num_classes*4 ,
|
969 |
+
dims=dims,
|
970 |
+
use_checkpoint=use_checkpoint,
|
971 |
+
use_scale_shift_norm=use_scale_shift_norm,
|
972 |
+
),
|
973 |
+
)
|
974 |
+
self._feature_size += ch
|
975 |
+
|
976 |
+
self.output_blocks = nn.ModuleList([])
|
977 |
+
for level, mult in list(enumerate(channel_mult))[::-1]:
|
978 |
+
for i in range(num_res_blocks + 1):
|
979 |
+
ich = input_block_chans.pop()
|
980 |
+
#print(ch, ich)
|
981 |
+
layers = [
|
982 |
+
SDMResBlock(
|
983 |
+
ch + ich,
|
984 |
+
time_embed_dim,
|
985 |
+
dropout,
|
986 |
+
c_channels=num_classes if mask_emb == "resize" else num_classes*4,
|
987 |
+
out_channels=int(model_channels * mult),
|
988 |
+
dims=dims,
|
989 |
+
use_checkpoint=use_checkpoint,
|
990 |
+
use_scale_shift_norm=use_scale_shift_norm,
|
991 |
+
SPADE_type=SPADE_type,
|
992 |
+
guidance_nc = ich,
|
993 |
+
debug=self.debug,
|
994 |
+
)
|
995 |
+
]
|
996 |
+
ch = int(model_channels * mult)
|
997 |
+
#print(ds)
|
998 |
+
if ds in attention_resolutions:
|
999 |
+
layers.append(
|
1000 |
+
AttentionBlock(
|
1001 |
+
ch,
|
1002 |
+
use_checkpoint=use_checkpoint,
|
1003 |
+
num_heads=num_heads_upsample,
|
1004 |
+
num_head_channels=num_head_channels,
|
1005 |
+
use_new_attention_order=use_new_attention_order,
|
1006 |
+
)
|
1007 |
+
)
|
1008 |
+
if level and i == num_res_blocks:
|
1009 |
+
out_ch = ch
|
1010 |
+
layers.append(
|
1011 |
+
SDMResBlock(
|
1012 |
+
ch,
|
1013 |
+
time_embed_dim,
|
1014 |
+
dropout,
|
1015 |
+
c_channels=num_classes if mask_emb == "resize" else num_classes*4,
|
1016 |
+
out_channels=out_ch,
|
1017 |
+
dims=dims,
|
1018 |
+
use_checkpoint=use_checkpoint,
|
1019 |
+
use_scale_shift_norm=use_scale_shift_norm,
|
1020 |
+
up=True,
|
1021 |
+
debug=self.debug
|
1022 |
+
)
|
1023 |
+
if resblock_updown
|
1024 |
+
else Upsample(ch, conv_resample, dims=dims, out_channels=out_ch)
|
1025 |
+
)
|
1026 |
+
ds //= 2
|
1027 |
+
self.output_blocks.append(TimestepEmbedSequential(*layers))
|
1028 |
+
self._feature_size += ch
|
1029 |
+
|
1030 |
+
self.out = nn.Sequential(
|
1031 |
+
normalization(ch),
|
1032 |
+
SiLU(),
|
1033 |
+
zero_module(conv_nd(dims, input_ch, out_channels, 3, padding=1)),
|
1034 |
+
)
|
1035 |
+
def _set_gradient_checkpointing(self, module, value=False):
|
1036 |
+
#if isinstance(module, (CrossAttnDownBlock2D, DownBlock2D, CrossAttnUpBlock2D, UpBlock2D)):
|
1037 |
+
module.gradient_checkpointing = value
|
1038 |
+
def forward(self, x, y=None, timesteps=None ):
|
1039 |
+
"""
|
1040 |
+
Apply the model to an input batch.
|
1041 |
+
|
1042 |
+
:param x: an [N x C x ...] Tensor of inputs.
|
1043 |
+
:param timesteps: a 1-D batch of timesteps.
|
1044 |
+
:param y: an [N] Tensor of labels, if class-conditional.
|
1045 |
+
:return: an [N x C x ...] Tensor of outputs.
|
1046 |
+
"""
|
1047 |
+
assert (y is not None) == (
|
1048 |
+
self.num_classes is not None
|
1049 |
+
), "must specify y if and only if the model is class-conditional"
|
1050 |
+
|
1051 |
+
hs = []
|
1052 |
+
if not th.is_tensor(timesteps):
|
1053 |
+
timesteps = th.tensor([timesteps], dtype=th.long, device=x.device)
|
1054 |
+
elif th.is_tensor(timesteps) and len(timesteps.shape) == 0:
|
1055 |
+
timesteps = timesteps[None].to(x.device)
|
1056 |
+
|
1057 |
+
timesteps = timestep_embedding(timesteps, self.model_channels).type(x.dtype).to(x.device)
|
1058 |
+
emb = self.time_embed(timesteps)
|
1059 |
+
|
1060 |
+
y = y.type(self.dtype)
|
1061 |
+
h = x.type(self.dtype)
|
1062 |
+
for module in self.input_blocks:
|
1063 |
+
# input_blocks have no any opts for y
|
1064 |
+
h = module(h, y, emb)
|
1065 |
+
#print(h.shape)
|
1066 |
+
hs.append(h)
|
1067 |
+
|
1068 |
+
h = self.middle_block(h, y, emb)
|
1069 |
+
|
1070 |
+
if self.debug:
|
1071 |
+
extra_list = []
|
1072 |
+
|
1073 |
+
for module in self.output_blocks:
|
1074 |
+
temp = hs.pop()
|
1075 |
+
|
1076 |
+
#print("before:", h.shape, temp.shape)
|
1077 |
+
# copy padding to match the downsample size
|
1078 |
+
if h.shape[2] != temp.shape[2]:
|
1079 |
+
p1d = (0, 0, 0, 1)
|
1080 |
+
h = F.pad(h, p1d, "replicate")
|
1081 |
+
|
1082 |
+
if h.shape[3] != temp.shape[3]:
|
1083 |
+
p2d = (0, 1, 0, 0)
|
1084 |
+
h = F.pad(h, p2d, "replicate")
|
1085 |
+
#print("after:", h.shape, temp.shape)
|
1086 |
+
|
1087 |
+
h = th.cat([h, temp], dim=1)
|
1088 |
+
if self.debug:
|
1089 |
+
h, extra = module(h, y, emb)
|
1090 |
+
extra_list.append(extra)
|
1091 |
+
else:
|
1092 |
+
h = module(h, y, emb)
|
1093 |
+
|
1094 |
+
h = h.type(x.dtype)
|
1095 |
+
|
1096 |
+
if not self.debug:
|
1097 |
+
return UNet2DOutput(sample=self.out(h))
|
1098 |
+
else:
|
1099 |
+
return UNet2DOutput(sample=self.out(h)), extra_list
|
1100 |
+
|
1101 |
+
|
1102 |
+
class SuperResModel(UNetModel):
|
1103 |
+
"""
|
1104 |
+
A UNetModel that performs super-resolution.
|
1105 |
+
|
1106 |
+
Expects an extra kwarg `low_res` to condition on a low-resolution image.
|
1107 |
+
"""
|
1108 |
+
|
1109 |
+
def __init__(self, image_size, in_channels, *args, **kwargs):
|
1110 |
+
super().__init__(image_size, in_channels * 2, *args, **kwargs)
|
1111 |
+
|
1112 |
+
def forward(self, x, cond, timesteps, low_res=None, **kwargs):
|
1113 |
+
_, _, new_height, new_width = x.shape
|
1114 |
+
upsampled = F.interpolate(low_res, (new_height, new_width), mode="bilinear")
|
1115 |
+
x = th.cat([x, upsampled], dim=1)
|
1116 |
+
return super().forward(x, cond, timesteps, **kwargs)
|
1117 |
+
|
1118 |
+
|
1119 |
+
class EncoderUNetModel(nn.Module):
|
1120 |
+
"""
|
1121 |
+
The half UNet model with attention and timestep embedding.
|
1122 |
+
|
1123 |
+
For usage, see UNet.
|
1124 |
+
"""
|
1125 |
+
|
1126 |
+
def __init__(
|
1127 |
+
self,
|
1128 |
+
image_size,
|
1129 |
+
in_channels,
|
1130 |
+
model_channels,
|
1131 |
+
out_channels,
|
1132 |
+
num_res_blocks,
|
1133 |
+
attention_resolutions,
|
1134 |
+
dropout=0,
|
1135 |
+
channel_mult=(1, 2, 4, 8),
|
1136 |
+
conv_resample=True,
|
1137 |
+
dims=2,
|
1138 |
+
use_checkpoint=False,
|
1139 |
+
use_fp16=False,
|
1140 |
+
num_heads=1,
|
1141 |
+
num_head_channels=-1,
|
1142 |
+
num_heads_upsample=-1,
|
1143 |
+
use_scale_shift_norm=False,
|
1144 |
+
resblock_updown=False,
|
1145 |
+
use_new_attention_order=False,
|
1146 |
+
pool="adaptive",
|
1147 |
+
):
|
1148 |
+
super().__init__()
|
1149 |
+
|
1150 |
+
if num_heads_upsample == -1:
|
1151 |
+
num_heads_upsample = num_heads
|
1152 |
+
|
1153 |
+
self.in_channels = in_channels
|
1154 |
+
self.model_channels = model_channels
|
1155 |
+
self.out_channels = out_channels
|
1156 |
+
self.num_res_blocks = num_res_blocks
|
1157 |
+
self.attention_resolutions = attention_resolutions
|
1158 |
+
self.dropout = dropout
|
1159 |
+
self.channel_mult = channel_mult
|
1160 |
+
self.conv_resample = conv_resample
|
1161 |
+
self.use_checkpoint = use_checkpoint
|
1162 |
+
self.num_heads = num_heads
|
1163 |
+
self.num_head_channels = num_head_channels
|
1164 |
+
self.num_heads_upsample = num_heads_upsample
|
1165 |
+
|
1166 |
+
time_embed_dim = model_channels * 4
|
1167 |
+
self.time_embed = nn.Sequential(
|
1168 |
+
linear(model_channels, time_embed_dim),
|
1169 |
+
SiLU(),
|
1170 |
+
linear(time_embed_dim, time_embed_dim),
|
1171 |
+
)
|
1172 |
+
|
1173 |
+
ch = int(channel_mult[0] * model_channels)
|
1174 |
+
self.input_blocks = nn.ModuleList(
|
1175 |
+
[TimestepEmbedSequential(conv_nd(dims, in_channels, ch, 3, padding=1))]
|
1176 |
+
)
|
1177 |
+
self._feature_size = ch
|
1178 |
+
input_block_chans = [ch]
|
1179 |
+
ds = 1
|
1180 |
+
for level, mult in enumerate(channel_mult):
|
1181 |
+
for _ in range(num_res_blocks):
|
1182 |
+
layers = [
|
1183 |
+
ResBlock(
|
1184 |
+
ch,
|
1185 |
+
time_embed_dim,
|
1186 |
+
dropout,
|
1187 |
+
out_channels=int(mult * model_channels),
|
1188 |
+
dims=dims,
|
1189 |
+
use_checkpoint=use_checkpoint,
|
1190 |
+
use_scale_shift_norm=use_scale_shift_norm,
|
1191 |
+
)
|
1192 |
+
]
|
1193 |
+
ch = int(mult * model_channels)
|
1194 |
+
if ds in attention_resolutions:
|
1195 |
+
layers.append(
|
1196 |
+
AttentionBlock(
|
1197 |
+
ch,
|
1198 |
+
use_checkpoint=use_checkpoint,
|
1199 |
+
num_heads=num_heads,
|
1200 |
+
num_head_channels=num_head_channels,
|
1201 |
+
use_new_attention_order=use_new_attention_order,
|
1202 |
+
)
|
1203 |
+
)
|
1204 |
+
self.input_blocks.append(TimestepEmbedSequential(*layers))
|
1205 |
+
self._feature_size += ch
|
1206 |
+
input_block_chans.append(ch)
|
1207 |
+
if level != len(channel_mult) - 1:
|
1208 |
+
out_ch = ch
|
1209 |
+
self.input_blocks.append(
|
1210 |
+
TimestepEmbedSequential(
|
1211 |
+
ResBlock(
|
1212 |
+
ch,
|
1213 |
+
time_embed_dim,
|
1214 |
+
dropout,
|
1215 |
+
out_channels=out_ch,
|
1216 |
+
dims=dims,
|
1217 |
+
use_checkpoint=use_checkpoint,
|
1218 |
+
use_scale_shift_norm=use_scale_shift_norm,
|
1219 |
+
down=True,
|
1220 |
+
)
|
1221 |
+
if resblock_updown
|
1222 |
+
else Downsample(
|
1223 |
+
ch, conv_resample, dims=dims, out_channels=out_ch
|
1224 |
+
)
|
1225 |
+
)
|
1226 |
+
)
|
1227 |
+
ch = out_ch
|
1228 |
+
input_block_chans.append(ch)
|
1229 |
+
ds *= 2
|
1230 |
+
self._feature_size += ch
|
1231 |
+
|
1232 |
+
self.middle_block = TimestepEmbedSequential(
|
1233 |
+
ResBlock(
|
1234 |
+
ch,
|
1235 |
+
time_embed_dim,
|
1236 |
+
dropout,
|
1237 |
+
dims=dims,
|
1238 |
+
use_checkpoint=use_checkpoint,
|
1239 |
+
use_scale_shift_norm=use_scale_shift_norm,
|
1240 |
+
),
|
1241 |
+
AttentionBlock(
|
1242 |
+
ch,
|
1243 |
+
use_checkpoint=use_checkpoint,
|
1244 |
+
num_heads=num_heads,
|
1245 |
+
num_head_channels=num_head_channels,
|
1246 |
+
use_new_attention_order=use_new_attention_order,
|
1247 |
+
),
|
1248 |
+
ResBlock(
|
1249 |
+
ch,
|
1250 |
+
time_embed_dim,
|
1251 |
+
dropout,
|
1252 |
+
dims=dims,
|
1253 |
+
use_checkpoint=use_checkpoint,
|
1254 |
+
use_scale_shift_norm=use_scale_shift_norm,
|
1255 |
+
),
|
1256 |
+
)
|
1257 |
+
self._feature_size += ch
|
1258 |
+
self.pool = pool
|
1259 |
+
if pool == "adaptive":
|
1260 |
+
self.out = nn.Sequential(
|
1261 |
+
normalization(ch),
|
1262 |
+
SiLU(),
|
1263 |
+
nn.AdaptiveAvgPool2d((1, 1)),
|
1264 |
+
zero_module(conv_nd(dims, ch, out_channels, 1)),
|
1265 |
+
nn.Flatten(),
|
1266 |
+
)
|
1267 |
+
elif pool == "attention":
|
1268 |
+
assert num_head_channels != -1
|
1269 |
+
self.out = nn.Sequential(
|
1270 |
+
normalization(ch),
|
1271 |
+
SiLU(),
|
1272 |
+
AttentionPool2d(
|
1273 |
+
(image_size // ds), ch, num_head_channels, out_channels
|
1274 |
+
),
|
1275 |
+
)
|
1276 |
+
elif pool == "spatial":
|
1277 |
+
self.out = nn.Sequential(
|
1278 |
+
nn.Linear(self._feature_size, 2048),
|
1279 |
+
nn.ReLU(),
|
1280 |
+
nn.Linear(2048, self.out_channels),
|
1281 |
+
)
|
1282 |
+
elif pool == "spatial_v2":
|
1283 |
+
self.out = nn.Sequential(
|
1284 |
+
nn.Linear(self._feature_size, 2048),
|
1285 |
+
normalization(2048),
|
1286 |
+
SiLU(),
|
1287 |
+
nn.Linear(2048, self.out_channels),
|
1288 |
+
)
|
1289 |
+
else:
|
1290 |
+
raise NotImplementedError(f"Unexpected {pool} pooling")
|
1291 |
+
def forward(self, x, timesteps):
|
1292 |
+
"""
|
1293 |
+
Apply the model to an input batch.
|
1294 |
+
|
1295 |
+
:param x: an [N x C x ...] Tensor of inputs.
|
1296 |
+
:param timesteps: a 1-D batch of timesteps.
|
1297 |
+
:return: an [N x K] Tensor of outputs.
|
1298 |
+
"""
|
1299 |
+
emb = self.time_embed(timestep_embedding(timesteps, self.model_channels))
|
1300 |
+
|
1301 |
+
results = []
|
1302 |
+
h = x.type(self.dtype)
|
1303 |
+
for module in self.input_blocks:
|
1304 |
+
h = module(h, emb)
|
1305 |
+
if self.pool.startswith("spatial"):
|
1306 |
+
results.append(h.type(x.dtype).mean(dim=(2, 3)))
|
1307 |
+
h = self.middle_block(h, emb)
|
1308 |
+
if self.pool.startswith("spatial"):
|
1309 |
+
results.append(h.type(x.dtype).mean(dim=(2, 3)))
|
1310 |
+
h = th.cat(results, axis=-1)
|
1311 |
+
return self.out(h)
|
1312 |
+
else:
|
1313 |
+
h = h.type(x.dtype)
|
1314 |
+
return self.out(h)
|
1315 |
+
|
diffusion_module/unet_2d_blocks.py
ADDED
The diff for this file is too large to render.
See raw diff
|
|
diffusion_module/unet_2d_sdm.py
ADDED
@@ -0,0 +1,357 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Copyright 2023 The HuggingFace Team. All rights reserved.
|
2 |
+
#
|
3 |
+
# Licensed under the Apache License, Version 2.0 (the "License");
|
4 |
+
# you may not use this file except in compliance with the License.
|
5 |
+
# You may obtain a copy of the License at
|
6 |
+
#
|
7 |
+
# http://www.apache.org/licenses/LICENSE-2.0
|
8 |
+
#
|
9 |
+
# Unless required by applicable law or agreed to in writing, software
|
10 |
+
# distributed under the License is distributed on an "AS IS" BASIS,
|
11 |
+
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
12 |
+
# See the License for the specific language governing permissions and
|
13 |
+
# limitations under the License.
|
14 |
+
from dataclasses import dataclass
|
15 |
+
from typing import Optional, Tuple, Union
|
16 |
+
|
17 |
+
import torch
|
18 |
+
import torch.nn as nn
|
19 |
+
|
20 |
+
from diffusers.configuration_utils import ConfigMixin, register_to_config
|
21 |
+
from diffusers.utils import BaseOutput
|
22 |
+
from diffusers.models.embeddings import GaussianFourierProjection, TimestepEmbedding, Timesteps
|
23 |
+
from diffusers.models.modeling_utils import ModelMixin
|
24 |
+
from .unet_2d_blocks import UNetSDMMidBlock2D, get_down_block, get_up_block, UNetSDMMidBlock2D
|
25 |
+
from diffusers.loaders import UNet2DConditionLoadersMixin
|
26 |
+
|
27 |
+
@dataclass
|
28 |
+
class UNet2DOutput(BaseOutput):
|
29 |
+
"""
|
30 |
+
Args:
|
31 |
+
sample (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)`):
|
32 |
+
Hidden states output. Output of last layer of model.
|
33 |
+
"""
|
34 |
+
|
35 |
+
sample: torch.FloatTensor
|
36 |
+
|
37 |
+
class SDMUNet2DModel(ModelMixin, ConfigMixin, UNet2DConditionLoadersMixin):
|
38 |
+
r"""
|
39 |
+
UNet2DModel is a 2D UNet model that takes in a noisy sample and a timestep and returns sample shaped output.
|
40 |
+
|
41 |
+
This model inherits from [`ModelMixin`]. Check the superclass documentation for the generic methods the library
|
42 |
+
implements for all the model (such as downloading or saving, etc.)
|
43 |
+
|
44 |
+
Parameters:
|
45 |
+
sample_size (`int` or `Tuple[int, int]`, *optional*, defaults to `None`):
|
46 |
+
Height and width of input/output sample. Dimensions must be a multiple of `2 ** (len(block_out_channels) -
|
47 |
+
1)`.
|
48 |
+
in_channels (`int`, *optional*, defaults to 3): Number of channels in the input image.
|
49 |
+
out_channels (`int`, *optional*, defaults to 3): Number of channels in the output.
|
50 |
+
center_input_sample (`bool`, *optional*, defaults to `False`): Whether to center the input sample.
|
51 |
+
time_embedding_type (`str`, *optional*, defaults to `"positional"`): Type of time embedding to use.
|
52 |
+
freq_shift (`int`, *optional*, defaults to 0): Frequency shift for fourier time embedding.
|
53 |
+
flip_sin_to_cos (`bool`, *optional*, defaults to :
|
54 |
+
obj:`True`): Whether to flip sin to cos for fourier time embedding.
|
55 |
+
down_block_types (`Tuple[str]`, *optional*, defaults to :
|
56 |
+
obj:`("DownBlock2D", "AttnDownBlock2D", "AttnDownBlock2D", "AttnDownBlock2D")`): Tuple of downsample block
|
57 |
+
types.
|
58 |
+
mid_block_type (`str`, *optional*, defaults to `"UNetMidBlock2D"`):
|
59 |
+
The mid block type. Choose from `UNetMidBlock2D` or `UnCLIPUNetMidBlock2D`.
|
60 |
+
up_block_types (`Tuple[str]`, *optional*, defaults to :
|
61 |
+
obj:`("AttnUpBlock2D", "AttnUpBlock2D", "AttnUpBlock2D", "UpBlock2D")`): Tuple of upsample block types.
|
62 |
+
block_out_channels (`Tuple[int]`, *optional*, defaults to :
|
63 |
+
obj:`(224, 448, 672, 896)`): Tuple of block output channels.
|
64 |
+
layers_per_block (`int`, *optional*, defaults to `2`): The number of layers per block.
|
65 |
+
mid_block_scale_factor (`float`, *optional*, defaults to `1`): The scale factor for the mid block.
|
66 |
+
downsample_padding (`int`, *optional*, defaults to `1`): The padding for the downsample convolution.
|
67 |
+
act_fn (`str`, *optional*, defaults to `"silu"`): The activation function to use.
|
68 |
+
attention_head_dim (`int`, *optional*, defaults to `8`): The attention head dimension.
|
69 |
+
norm_num_groups (`int`, *optional*, defaults to `32`): The number of groups for the normalization.
|
70 |
+
norm_eps (`float`, *optional*, defaults to `1e-5`): The epsilon for the normalization.
|
71 |
+
resnet_time_scale_shift (`str`, *optional*, defaults to `"default"`): Time scale shift config
|
72 |
+
for resnet blocks, see [`~models.resnet.ResnetBlock2D`]. Choose from `default` or `scale_shift`.
|
73 |
+
class_embed_type (`str`, *optional*, defaults to None):
|
74 |
+
The type of class embedding to use which is ultimately summed with the time embeddings. Choose from `None`,
|
75 |
+
`"timestep"`, or `"identity"`.
|
76 |
+
num_class_embeds (`int`, *optional*, defaults to None):
|
77 |
+
Input dimension of the learnable embedding matrix to be projected to `time_embed_dim`, when performing
|
78 |
+
class conditioning with `class_embed_type` equal to `None`.
|
79 |
+
"""
|
80 |
+
|
81 |
+
_supports_gradient_checkpointing = True
|
82 |
+
@register_to_config
|
83 |
+
def __init__(
|
84 |
+
self,
|
85 |
+
sample_size: Optional[Union[int, Tuple[int, int]]] = None,
|
86 |
+
in_channels: int = 3,
|
87 |
+
out_channels: int = 3,
|
88 |
+
center_input_sample: bool = False,
|
89 |
+
time_embedding_type: str = "positional",
|
90 |
+
freq_shift: int = 0,
|
91 |
+
flip_sin_to_cos: bool = True,
|
92 |
+
down_block_types: Tuple[str] = ("DownBlock2D", "AttnDownBlock2D", "AttnDownBlock2D", "AttnDownBlock2D"),
|
93 |
+
up_block_types: Tuple[str] = ("AttnUpBlock2D", "AttnUpBlock2D", "AttnUpBlock2D", "UpBlock2D"),
|
94 |
+
block_out_channels: Tuple[int] = (224, 448, 672, 896),
|
95 |
+
layers_per_block: int = 2,
|
96 |
+
mid_block_scale_factor: float = 1,
|
97 |
+
downsample_padding: int = 1,
|
98 |
+
act_fn: str = "silu",
|
99 |
+
attention_head_dim: Optional[int] = 8,
|
100 |
+
norm_num_groups: int = 32,
|
101 |
+
norm_eps: float = 1e-5,
|
102 |
+
resnet_time_scale_shift: str = "scale_shift",
|
103 |
+
add_attention: bool = True,
|
104 |
+
class_embed_type: Optional[str] = None,
|
105 |
+
num_class_embeds: Optional[int] = None,
|
106 |
+
segmap_channels: int = 34,
|
107 |
+
):
|
108 |
+
super().__init__()
|
109 |
+
|
110 |
+
self.sample_size = sample_size
|
111 |
+
self.segmap_channels = segmap_channels
|
112 |
+
time_embed_dim = block_out_channels[0] * 4
|
113 |
+
|
114 |
+
# Check inputs
|
115 |
+
if len(down_block_types) != len(up_block_types):
|
116 |
+
raise ValueError(
|
117 |
+
f"Must provide the same number of `down_block_types` as `up_block_types`. `down_block_types`: {down_block_types}. `up_block_types`: {up_block_types}."
|
118 |
+
)
|
119 |
+
|
120 |
+
if len(block_out_channels) != len(down_block_types):
|
121 |
+
raise ValueError(
|
122 |
+
f"Must provide the same number of `block_out_channels` as `down_block_types`. `block_out_channels`: {block_out_channels}. `down_block_types`: {down_block_types}."
|
123 |
+
)
|
124 |
+
|
125 |
+
# input
|
126 |
+
self.conv_in = nn.Conv2d(in_channels, block_out_channels[0], kernel_size=3, padding=(1, 1))
|
127 |
+
|
128 |
+
# time
|
129 |
+
if time_embedding_type == "fourier":
|
130 |
+
self.time_proj = GaussianFourierProjection(embedding_size=block_out_channels[0], scale=16)
|
131 |
+
timestep_input_dim = 2 * block_out_channels[0]
|
132 |
+
elif time_embedding_type == "positional":
|
133 |
+
self.time_proj = Timesteps(block_out_channels[0], flip_sin_to_cos, freq_shift)
|
134 |
+
timestep_input_dim = block_out_channels[0]
|
135 |
+
|
136 |
+
self.time_embedding = TimestepEmbedding(timestep_input_dim, time_embed_dim)
|
137 |
+
|
138 |
+
# class embedding
|
139 |
+
if class_embed_type is None and num_class_embeds is not None:
|
140 |
+
self.class_embedding = nn.Embedding(num_class_embeds, time_embed_dim)
|
141 |
+
elif class_embed_type == "timestep":
|
142 |
+
self.class_embedding = TimestepEmbedding(timestep_input_dim, time_embed_dim)
|
143 |
+
elif class_embed_type == "identity":
|
144 |
+
self.class_embedding = nn.Identity(time_embed_dim, time_embed_dim)
|
145 |
+
else:
|
146 |
+
self.class_embedding = None
|
147 |
+
|
148 |
+
self.down_blocks = nn.ModuleList([])
|
149 |
+
self.mid_block = None
|
150 |
+
self.up_blocks = nn.ModuleList([])
|
151 |
+
|
152 |
+
# down
|
153 |
+
output_channel = block_out_channels[0]
|
154 |
+
for i, down_block_type in enumerate(down_block_types):
|
155 |
+
input_channel = output_channel
|
156 |
+
output_channel = block_out_channels[i]
|
157 |
+
is_final_block = i == len(block_out_channels) - 1
|
158 |
+
|
159 |
+
down_block = get_down_block(
|
160 |
+
down_block_type,
|
161 |
+
num_layers=layers_per_block,
|
162 |
+
in_channels=input_channel,
|
163 |
+
out_channels=output_channel,
|
164 |
+
temb_channels=time_embed_dim,
|
165 |
+
add_downsample=not is_final_block,
|
166 |
+
resnet_eps=norm_eps,
|
167 |
+
resnet_act_fn=act_fn,
|
168 |
+
resnet_groups=norm_num_groups,
|
169 |
+
attn_num_head_channels=attention_head_dim,
|
170 |
+
downsample_padding=downsample_padding,
|
171 |
+
resnet_time_scale_shift=resnet_time_scale_shift,
|
172 |
+
)
|
173 |
+
self.down_blocks.append(down_block)
|
174 |
+
|
175 |
+
# mid
|
176 |
+
self.mid_block = UNetSDMMidBlock2D(
|
177 |
+
in_channels=block_out_channels[-1],
|
178 |
+
temb_channels=time_embed_dim,
|
179 |
+
resnet_eps=norm_eps,
|
180 |
+
resnet_act_fn=act_fn,
|
181 |
+
output_scale_factor=mid_block_scale_factor,
|
182 |
+
resnet_time_scale_shift=resnet_time_scale_shift,
|
183 |
+
attn_num_head_channels=attention_head_dim,
|
184 |
+
resnet_groups=norm_num_groups,
|
185 |
+
add_attention=add_attention,
|
186 |
+
segmap_channels=segmap_channels,
|
187 |
+
)
|
188 |
+
|
189 |
+
# up
|
190 |
+
reversed_block_out_channels = list(reversed(block_out_channels))
|
191 |
+
output_channel = reversed_block_out_channels[0]
|
192 |
+
for i, up_block_type in enumerate(up_block_types):
|
193 |
+
prev_output_channel = output_channel
|
194 |
+
output_channel = reversed_block_out_channels[i]
|
195 |
+
input_channel = reversed_block_out_channels[min(i + 1, len(block_out_channels) - 1)]
|
196 |
+
|
197 |
+
is_final_block = i == len(block_out_channels) - 1
|
198 |
+
|
199 |
+
up_block = get_up_block(
|
200 |
+
up_block_type,
|
201 |
+
num_layers=layers_per_block + 1,
|
202 |
+
in_channels=input_channel,
|
203 |
+
out_channels=output_channel,
|
204 |
+
prev_output_channel=prev_output_channel,
|
205 |
+
temb_channels=time_embed_dim,
|
206 |
+
add_upsample=not is_final_block,
|
207 |
+
resnet_eps=norm_eps,
|
208 |
+
resnet_act_fn=act_fn,
|
209 |
+
resnet_groups=norm_num_groups,
|
210 |
+
attn_num_head_channels=attention_head_dim,
|
211 |
+
resnet_time_scale_shift=resnet_time_scale_shift,
|
212 |
+
segmap_channels=segmap_channels,
|
213 |
+
)
|
214 |
+
self.up_blocks.append(up_block)
|
215 |
+
prev_output_channel = output_channel
|
216 |
+
|
217 |
+
# out
|
218 |
+
num_groups_out = norm_num_groups if norm_num_groups is not None else min(block_out_channels[0] // 4, 32)
|
219 |
+
self.conv_norm_out = nn.GroupNorm(num_channels=block_out_channels[0], num_groups=num_groups_out, eps=norm_eps)
|
220 |
+
self.conv_act = nn.SiLU()
|
221 |
+
self.conv_out = nn.Conv2d(block_out_channels[0], out_channels, kernel_size=3, padding=1)
|
222 |
+
def _set_gradient_checkpointing(self, module, value=False):
|
223 |
+
#if isinstance(module, (CrossAttnDownBlock2D, DownBlock2D, CrossAttnUpBlock2D, UpBlock2D)):
|
224 |
+
module.gradient_checkpointing = value
|
225 |
+
def forward(
|
226 |
+
self,
|
227 |
+
sample: torch.FloatTensor,
|
228 |
+
segmap: torch.FloatTensor,
|
229 |
+
timestep: Union[torch.Tensor, float, int],
|
230 |
+
class_labels: Optional[torch.Tensor] = None,
|
231 |
+
return_dict: bool = True,
|
232 |
+
) -> Union[UNet2DOutput, Tuple]:
|
233 |
+
r"""
|
234 |
+
Args:
|
235 |
+
sample (`torch.FloatTensor`): (batch, channel, height, width) noisy inputs tensor
|
236 |
+
timestep (`torch.FloatTensor` or `float` or `int): (batch) timesteps
|
237 |
+
class_labels (`torch.FloatTensor`, *optional*, defaults to `None`):
|
238 |
+
Optional class labels for conditioning. Their embeddings will be summed with the timestep embeddings.
|
239 |
+
return_dict (`bool`, *optional*, defaults to `True`):
|
240 |
+
Whether or not to return a [`~models.unet_2d.UNet2DOutput`] instead of a plain tuple.
|
241 |
+
|
242 |
+
Returns:
|
243 |
+
[`~models.unet_2d.UNet2DOutput`] or `tuple`: [`~models.unet_2d.UNet2DOutput`] if `return_dict` is True,
|
244 |
+
otherwise a `tuple`. When returning a tuple, the first element is the sample tensor.
|
245 |
+
"""
|
246 |
+
# 0. center input if necessary
|
247 |
+
if self.config.center_input_sample:
|
248 |
+
sample = 2 * sample - 1.0
|
249 |
+
|
250 |
+
# 1. time
|
251 |
+
#print(timestep.shape)
|
252 |
+
timesteps = timestep
|
253 |
+
if not torch.is_tensor(timesteps):
|
254 |
+
timesteps = torch.tensor([timesteps], dtype=torch.long, device=sample.device)
|
255 |
+
elif torch.is_tensor(timesteps) and len(timesteps.shape) == 0:
|
256 |
+
timesteps = timesteps[None].to(sample.device)
|
257 |
+
|
258 |
+
# broadcast to batch dimension in a way that's compatible with ONNX/Core ML
|
259 |
+
timesteps = timesteps * torch.ones(sample.shape[0], dtype=timesteps.dtype, device=timesteps.device)
|
260 |
+
|
261 |
+
t_emb = self.time_proj(timesteps)
|
262 |
+
|
263 |
+
# timesteps does not contain any weights and will always return f32 tensors
|
264 |
+
# but time_embedding might actually be running in fp16. so we need to cast here.
|
265 |
+
# there might be better ways to encapsulate this.
|
266 |
+
t_emb = t_emb.to(dtype=self.dtype)
|
267 |
+
emb = self.time_embedding(t_emb)
|
268 |
+
|
269 |
+
if self.class_embedding is not None:
|
270 |
+
if class_labels is None:
|
271 |
+
raise ValueError("class_labels should be provided when doing class conditioning")
|
272 |
+
|
273 |
+
if self.config.class_embed_type == "timestep":
|
274 |
+
class_labels = self.time_proj(class_labels)
|
275 |
+
|
276 |
+
class_emb = self.class_embedding(class_labels).to(dtype=self.dtype)
|
277 |
+
emb = emb + class_emb
|
278 |
+
|
279 |
+
# 2. pre-process
|
280 |
+
skip_sample = sample
|
281 |
+
sample = self.conv_in(sample)
|
282 |
+
|
283 |
+
# 3. down
|
284 |
+
down_block_res_samples = (sample,)
|
285 |
+
for downsample_block in self.down_blocks:
|
286 |
+
if hasattr(downsample_block, "skip_conv"):
|
287 |
+
sample, res_samples, skip_sample = downsample_block(
|
288 |
+
hidden_states=sample, temb=emb, skip_sample=skip_sample,segmap=segmap
|
289 |
+
)
|
290 |
+
else:
|
291 |
+
sample, res_samples = downsample_block(hidden_states=sample, temb=emb)
|
292 |
+
|
293 |
+
down_block_res_samples += res_samples
|
294 |
+
|
295 |
+
# 4. mid
|
296 |
+
sample = self.mid_block(sample, segmap, emb)
|
297 |
+
|
298 |
+
# 5. up
|
299 |
+
skip_sample = None
|
300 |
+
for upsample_block in self.up_blocks:
|
301 |
+
res_samples = down_block_res_samples[-len(upsample_block.resnets) :]
|
302 |
+
down_block_res_samples = down_block_res_samples[: -len(upsample_block.resnets)]
|
303 |
+
|
304 |
+
if hasattr(upsample_block, "skip_conv"):
|
305 |
+
sample, skip_sample = upsample_block(sample, segmap, res_samples, emb, skip_sample)
|
306 |
+
else:
|
307 |
+
sample = upsample_block(sample, segmap, res_samples, emb)
|
308 |
+
|
309 |
+
# 6. post-process
|
310 |
+
sample = self.conv_norm_out(sample)
|
311 |
+
sample = self.conv_act(sample)
|
312 |
+
sample = self.conv_out(sample)
|
313 |
+
|
314 |
+
if skip_sample is not None:
|
315 |
+
sample += skip_sample
|
316 |
+
|
317 |
+
if self.config.time_embedding_type == "fourier":
|
318 |
+
timesteps = timesteps.reshape((sample.shape[0], *([1] * len(sample.shape[1:]))))
|
319 |
+
sample = sample / timesteps
|
320 |
+
|
321 |
+
if not return_dict:
|
322 |
+
return (sample,)
|
323 |
+
|
324 |
+
return UNet2DOutput(sample=sample)
|
325 |
+
|
326 |
+
|
327 |
+
if __name__ == "__main__":
|
328 |
+
path = 'output.txt'
|
329 |
+
f = open(path, 'w')
|
330 |
+
|
331 |
+
unet = SDMUNet2DModel(
|
332 |
+
sample_size=270,
|
333 |
+
in_channels=3,
|
334 |
+
out_channels=3,
|
335 |
+
layers_per_block=2,
|
336 |
+
block_out_channels=(256, 256, 512, 1024, 1024),
|
337 |
+
down_block_types=(
|
338 |
+
"ResnetDownsampleBlock2D",
|
339 |
+
"ResnetDownsampleBlock2D",
|
340 |
+
"ResnetDownsampleBlock2D",
|
341 |
+
"AttnDownBlock2D",
|
342 |
+
"AttnDownBlock2D",
|
343 |
+
),
|
344 |
+
up_block_types=(
|
345 |
+
"SDMAttnUpBlock2D",
|
346 |
+
"SDMAttnUpBlock2D",
|
347 |
+
"SDMResnetUpsampleBlock2D",
|
348 |
+
"SDMResnetUpsampleBlock2D",
|
349 |
+
"SDMResnetUpsampleBlock2D",
|
350 |
+
),
|
351 |
+
segmap_channels=34+1
|
352 |
+
)
|
353 |
+
|
354 |
+
print(unet,file=f)
|
355 |
+
f.close()
|
356 |
+
|
357 |
+
#summary(unet, [(1, 3, 270, 360), (1, 3, 270, 360), (2,)], device="cpu")
|
diffusion_module/utils/LSDMPipeline_expandDataset.py
ADDED
@@ -0,0 +1,179 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import inspect
|
2 |
+
from typing import List, Optional, Tuple, Union
|
3 |
+
|
4 |
+
import torch
|
5 |
+
|
6 |
+
from diffusers.models import UNet2DModel, VQModel
|
7 |
+
from diffusers.schedulers import DDIMScheduler
|
8 |
+
from diffusers.utils import randn_tensor
|
9 |
+
from diffusers.pipeline_utils import DiffusionPipeline, ImagePipelineOutput
|
10 |
+
import copy
|
11 |
+
|
12 |
+
class SDMLDMPipeline(DiffusionPipeline):
|
13 |
+
r"""
|
14 |
+
This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the
|
15 |
+
library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)
|
16 |
+
|
17 |
+
Parameters:
|
18 |
+
vae ([`VQModel`]):
|
19 |
+
Vector-quantized (VQ) Model to encode and decode images to and from latent representations.
|
20 |
+
unet ([`UNet2DModel`]): U-Net architecture to denoise the encoded image latents.
|
21 |
+
scheduler ([`SchedulerMixin`]):
|
22 |
+
[`DDIMScheduler`] is to be used in combination with `unet` to denoise the encoded image latents.
|
23 |
+
"""
|
24 |
+
|
25 |
+
def __init__(self, vae: VQModel, unet: UNet2DModel, scheduler: DDIMScheduler, torch_dtype=torch.float16, resolution=512, resolution_type="city"):
|
26 |
+
super().__init__()
|
27 |
+
self.register_modules(vae=vae, unet=unet, scheduler=scheduler)
|
28 |
+
self.torch_dtype = torch_dtype
|
29 |
+
self.resolution = resolution
|
30 |
+
self.resolution_type = resolution_type
|
31 |
+
@torch.no_grad()
|
32 |
+
def __call__(
|
33 |
+
self,
|
34 |
+
segmap = None,
|
35 |
+
batch_size: int = 8,
|
36 |
+
generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
|
37 |
+
eta: float = 0.0,
|
38 |
+
num_inference_steps: int = 1000,
|
39 |
+
output_type: Optional[str] = "pil",
|
40 |
+
return_dict: bool = True,
|
41 |
+
every_step_save: int = None,
|
42 |
+
s: int = 1,
|
43 |
+
num_evolution_per_mask = 10,
|
44 |
+
debug = False,
|
45 |
+
**kwargs,
|
46 |
+
) -> Union[Tuple, ImagePipelineOutput]:
|
47 |
+
r"""
|
48 |
+
Args:
|
49 |
+
batch_size (`int`, *optional*, defaults to 1):
|
50 |
+
Number of images to generate.
|
51 |
+
generator (`torch.Generator`, *optional*):
|
52 |
+
One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
|
53 |
+
to make generation deterministic.
|
54 |
+
num_inference_steps (`int`, *optional*, defaults to 50):
|
55 |
+
The number of denoising steps. More denoising steps usually lead to a higher quality image at the
|
56 |
+
expense of slower inference.
|
57 |
+
output_type (`str`, *optional*, defaults to `"pil"`):
|
58 |
+
The output format of the generate image. Choose between
|
59 |
+
[PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
|
60 |
+
return_dict (`bool`, *optional*, defaults to `True`):
|
61 |
+
Whether or not to return a [`~pipelines.ImagePipelineOutput`] instead of a plain tuple.
|
62 |
+
|
63 |
+
Returns:
|
64 |
+
[`~pipelines.ImagePipelineOutput`] or `tuple`: [`~pipelines.model.ImagePipelineOutput`] if `return_dict` is
|
65 |
+
True, otherwise a `tuple. When returning a tuple, the first element is a list with the generated images.
|
66 |
+
"""
|
67 |
+
# self.unet.config.sample_size = (64, 64) # (135,180)
|
68 |
+
# self.unet.config.sample_size = (135,180)
|
69 |
+
if self.resolution_type == "crack":
|
70 |
+
self.unet.config.sample_size = (64,64)
|
71 |
+
elif self.resolution_type == "crack_256":
|
72 |
+
self.unet.config.sample_size = (256,256)
|
73 |
+
else:
|
74 |
+
sc = 1080 // self.resolution
|
75 |
+
latent_size = (self.resolution // 4, 1440 // (sc*4))
|
76 |
+
self.unet.config.sample_size = latent_size
|
77 |
+
#
|
78 |
+
if not isinstance(self.unet.config.sample_size, tuple):
|
79 |
+
self.unet.config.sample_size = (self.unet.config.sample_size, self.unet.config.sample_size)
|
80 |
+
|
81 |
+
if segmap is None:
|
82 |
+
print("Didn't inpute any segmap, use the empty as the input")
|
83 |
+
segmap = torch.zeros(batch_size,self.unet.config.segmap_channels, self.unet.config.sample_size[0], self.unet.config.sample_size[1])
|
84 |
+
segmap = segmap.to(self.device).type(self.torch_dtype)
|
85 |
+
if batch_size == 1 and num_evolution_per_mask > batch_size:
|
86 |
+
latents = randn_tensor(
|
87 |
+
(num_evolution_per_mask, self.unet.config.in_channels, self.unet.config.sample_size[0], self.unet.config.sample_size[1]),
|
88 |
+
generator=generator,
|
89 |
+
)
|
90 |
+
else:
|
91 |
+
latents = randn_tensor(
|
92 |
+
(batch_size, self.unet.config.in_channels, self.unet.config.sample_size[0], self.unet.config.sample_size[1]),
|
93 |
+
generator=generator,
|
94 |
+
)
|
95 |
+
latents = latents.to(self.device).type(self.torch_dtype)
|
96 |
+
|
97 |
+
# scale the initial noise by the standard deviation required by the scheduler (need to check)
|
98 |
+
latents = latents * self.scheduler.init_noise_sigma
|
99 |
+
|
100 |
+
self.scheduler.set_timesteps(num_inference_steps=num_inference_steps)
|
101 |
+
|
102 |
+
# prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
|
103 |
+
accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
|
104 |
+
|
105 |
+
extra_kwargs = {}
|
106 |
+
if accepts_eta:
|
107 |
+
extra_kwargs["eta"] = eta
|
108 |
+
|
109 |
+
step_latent = []
|
110 |
+
learn_sigma = True if hasattr(self.scheduler, "variance_type") else False
|
111 |
+
if debug:
|
112 |
+
extra_list_list = []
|
113 |
+
self.unet.debug=True
|
114 |
+
for i, t in enumerate(self.progress_bar(self.scheduler.timesteps)):
|
115 |
+
|
116 |
+
latent_model_input = self.scheduler.scale_model_input(latents, t)
|
117 |
+
# predict the noise residual
|
118 |
+
if debug:
|
119 |
+
output, extra_list = self.unet(latent_model_input, segmap, t)
|
120 |
+
noise_prediction = output.sample
|
121 |
+
extra_list_list.append(extra_list)
|
122 |
+
else:
|
123 |
+
noise_prediction = self.unet(latent_model_input, segmap, t).sample
|
124 |
+
# compute the previous noisy sample x_t -> x_t-1
|
125 |
+
|
126 |
+
|
127 |
+
if learn_sigma and "learn" in self.scheduler.variance_type:
|
128 |
+
model_pred, var_pred = torch.split(noise_prediction, latents.shape[1], dim=1)
|
129 |
+
else:
|
130 |
+
model_pred = noise_prediction
|
131 |
+
if s > 1.0:
|
132 |
+
if debug:
|
133 |
+
model_output_zero = self.unet(latent_model_input, torch.zeros_like(segmap), t)[0].sample
|
134 |
+
else:
|
135 |
+
model_output_zero = self.unet(latent_model_input, torch.zeros_like(segmap), t).sample
|
136 |
+
if learn_sigma and "learn" in self.scheduler.variance_type:
|
137 |
+
model_output_zero,_ = torch.split(model_output_zero, latents.shape[1], dim=1)
|
138 |
+
model_pred = model_pred + s * (model_pred - model_output_zero)
|
139 |
+
if learn_sigma and "learn" in self.scheduler.variance_type:
|
140 |
+
recombined = torch.cat((model_pred, var_pred), dim=1)
|
141 |
+
# when apply different scheduler, mean only !!
|
142 |
+
if learn_sigma and "learn" in self.scheduler.variance_type:
|
143 |
+
latents = self.scheduler.step(recombined, t, latents, **extra_kwargs).prev_sample
|
144 |
+
else:
|
145 |
+
latents = self.scheduler.step(noise_prediction, t, latents, **extra_kwargs).prev_sample
|
146 |
+
|
147 |
+
if every_step_save is not None:
|
148 |
+
if (i+1) % every_step_save == 0:
|
149 |
+
step_latent.append(copy.deepcopy(latents))
|
150 |
+
|
151 |
+
if debug:
|
152 |
+
return extra_list_list[-1]
|
153 |
+
|
154 |
+
# decode the image latents with the VAE
|
155 |
+
if every_step_save is not None:
|
156 |
+
image = []
|
157 |
+
for i, l in enumerate(step_latent):
|
158 |
+
l /= self.vae.config.scaling_factor # (0.18215)
|
159 |
+
#latents /= 7.706491063029163
|
160 |
+
l = self.vae.decode(l, segmap)
|
161 |
+
l = (l / 2 + 0.5).clamp(0, 1)
|
162 |
+
l = l.cpu().permute(0, 2, 3, 1).numpy()
|
163 |
+
if output_type == "pil":
|
164 |
+
l = self.numpy_to_pil(l)
|
165 |
+
image.append(l)
|
166 |
+
else:
|
167 |
+
latents /= self.vae.config.scaling_factor#(0.18215)
|
168 |
+
#latents /= 7.706491063029163
|
169 |
+
# image = self.vae.decode(latents, segmap).sample
|
170 |
+
image = self.vae.decode(latents, return_dict=False)[0]
|
171 |
+
image = (image / 2 + 0.5).clamp(0, 1)
|
172 |
+
image = image.cpu().permute(0, 2, 3, 1).numpy()
|
173 |
+
if output_type == "pil":
|
174 |
+
image = self.numpy_to_pil(image)
|
175 |
+
|
176 |
+
if not return_dict:
|
177 |
+
return (image,)
|
178 |
+
|
179 |
+
return ImagePipelineOutput(images=image)
|
diffusion_module/utils/Pipline.py
ADDED
@@ -0,0 +1,361 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import inspect
|
2 |
+
from typing import List, Optional, Tuple, Union
|
3 |
+
|
4 |
+
import torch
|
5 |
+
|
6 |
+
from diffusers.models import UNet2DModel, VQModel
|
7 |
+
from diffusers.schedulers import DDIMScheduler
|
8 |
+
from diffusers.utils import randn_tensor
|
9 |
+
from diffusers.pipeline_utils import DiffusionPipeline, ImagePipelineOutput
|
10 |
+
import copy
|
11 |
+
|
12 |
+
class LDMPipeline(DiffusionPipeline):
|
13 |
+
r"""
|
14 |
+
This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the
|
15 |
+
library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)
|
16 |
+
|
17 |
+
Parameters:
|
18 |
+
vae ([`VQModel`]):
|
19 |
+
Vector-quantized (VQ) Model to encode and decode images to and from latent representations.
|
20 |
+
unet ([`UNet2DModel`]): U-Net architecture to denoise the encoded image latents.
|
21 |
+
scheduler ([`SchedulerMixin`]):
|
22 |
+
[`DDIMScheduler`] is to be used in combination with `unet` to denoise the encoded image latents.
|
23 |
+
"""
|
24 |
+
|
25 |
+
def __init__(self, vae: VQModel, unet: UNet2DModel, scheduler: DDIMScheduler, torch_dtype=torch.float16):
|
26 |
+
super().__init__()
|
27 |
+
self.register_modules(vae=vae, unet=unet, scheduler=scheduler)
|
28 |
+
self.torch_dtype = torch_dtype
|
29 |
+
|
30 |
+
@torch.no_grad()
|
31 |
+
def __call__(
|
32 |
+
self,
|
33 |
+
batch_size: int = 8,
|
34 |
+
generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
|
35 |
+
eta: float = 0.0,
|
36 |
+
num_inference_steps: int = 1000,
|
37 |
+
output_type: Optional[str] = "pil",
|
38 |
+
return_dict: bool = True,
|
39 |
+
**kwargs,
|
40 |
+
) -> Union[Tuple, ImagePipelineOutput]:
|
41 |
+
r"""
|
42 |
+
Args:
|
43 |
+
batch_size (`int`, *optional*, defaults to 1):
|
44 |
+
Number of images to generate.
|
45 |
+
generator (`torch.Generator`, *optional*):
|
46 |
+
One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
|
47 |
+
to make generation deterministic.
|
48 |
+
num_inference_steps (`int`, *optional*, defaults to 50):
|
49 |
+
The number of denoising steps. More denoising steps usually lead to a higher quality image at the
|
50 |
+
expense of slower inference.
|
51 |
+
output_type (`str`, *optional*, defaults to `"pil"`):
|
52 |
+
The output format of the generate image. Choose between
|
53 |
+
[PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
|
54 |
+
return_dict (`bool`, *optional*, defaults to `True`):
|
55 |
+
Whether or not to return a [`~pipelines.ImagePipelineOutput`] instead of a plain tuple.
|
56 |
+
|
57 |
+
Returns:
|
58 |
+
[`~pipelines.ImagePipelineOutput`] or `tuple`: [`~pipelines.model.ImagePipelineOutput`] if `return_dict` is
|
59 |
+
True, otherwise a `tuple. When returning a tuple, the first element is a list with the generated images.
|
60 |
+
"""
|
61 |
+
if not isinstance(self.unet.config.sample_size,tuple):
|
62 |
+
self.unet.config.sample_size = (self.unet.config.sample_size,self.unet.config.sample_size)
|
63 |
+
|
64 |
+
latents = randn_tensor(
|
65 |
+
(batch_size, self.unet.config.in_channels, self.unet.config.sample_size[0], self.unet.config.sample_size[1]),
|
66 |
+
generator=generator,
|
67 |
+
)
|
68 |
+
latents = latents.to(self.device).type(self.torch_dtype)
|
69 |
+
|
70 |
+
# scale the initial noise by the standard deviation required by the scheduler (need to check)
|
71 |
+
latents = latents * self.scheduler.init_noise_sigma
|
72 |
+
|
73 |
+
self.scheduler.set_timesteps(num_inference_steps)
|
74 |
+
|
75 |
+
# prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
|
76 |
+
accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
|
77 |
+
|
78 |
+
extra_kwargs = {}
|
79 |
+
if accepts_eta:
|
80 |
+
extra_kwargs["eta"] = eta
|
81 |
+
|
82 |
+
for t in self.progress_bar(self.scheduler.timesteps):
|
83 |
+
latent_model_input = self.scheduler.scale_model_input(latents, t)
|
84 |
+
# predict the noise residual
|
85 |
+
noise_prediction = self.unet(latent_model_input, t).sample
|
86 |
+
# compute the previous noisy sample x_t -> x_t-1
|
87 |
+
latents = self.scheduler.step(noise_prediction, t, latents, **extra_kwargs).prev_sample
|
88 |
+
|
89 |
+
# decode the image latents with the VAE
|
90 |
+
latents /= self.vae.config.scaling_factor#(0.18215)
|
91 |
+
image = self.vae.decode(latents).sample
|
92 |
+
|
93 |
+
image = (image / 2 + 0.5).clamp(0, 1)
|
94 |
+
image = image.cpu().permute(0, 2, 3, 1).numpy()
|
95 |
+
if output_type == "pil":
|
96 |
+
image = self.numpy_to_pil(image)
|
97 |
+
|
98 |
+
if not return_dict:
|
99 |
+
return (image,)
|
100 |
+
|
101 |
+
return ImagePipelineOutput(images=image)
|
102 |
+
|
103 |
+
|
104 |
+
class SDMLDMPipeline(DiffusionPipeline):
|
105 |
+
r"""
|
106 |
+
This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the
|
107 |
+
library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)
|
108 |
+
|
109 |
+
Parameters:
|
110 |
+
vae ([`VQModel`]):
|
111 |
+
Vector-quantized (VQ) Model to encode and decode images to and from latent representations.
|
112 |
+
unet ([`UNet2DModel`]): U-Net architecture to denoise the encoded image latents.
|
113 |
+
scheduler ([`SchedulerMixin`]):
|
114 |
+
[`DDIMScheduler`] is to be used in combination with `unet` to denoise the encoded image latents.
|
115 |
+
"""
|
116 |
+
|
117 |
+
def __init__(self, vae: VQModel, unet: UNet2DModel, scheduler: DDIMScheduler, torch_dtype=torch.float16, resolution=512, resolution_type="city"):
|
118 |
+
super().__init__()
|
119 |
+
self.register_modules(vae=vae, unet=unet, scheduler=scheduler)
|
120 |
+
self.torch_dtype = torch_dtype
|
121 |
+
self.resolution = resolution
|
122 |
+
self.resolution_type = resolution_type
|
123 |
+
@torch.no_grad()
|
124 |
+
def __call__(
|
125 |
+
self,
|
126 |
+
segmap = None,
|
127 |
+
batch_size: int = 8,
|
128 |
+
generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
|
129 |
+
eta: float = 0.0,
|
130 |
+
num_inference_steps: int = 1000,
|
131 |
+
output_type: Optional[str] = "pil",
|
132 |
+
return_dict: bool = True,
|
133 |
+
every_step_save: int = None,
|
134 |
+
s: int = 1,
|
135 |
+
**kwargs,
|
136 |
+
) -> Union[Tuple, ImagePipelineOutput]:
|
137 |
+
r"""
|
138 |
+
Args:
|
139 |
+
batch_size (`int`, *optional*, defaults to 1):
|
140 |
+
Number of images to generate.
|
141 |
+
generator (`torch.Generator`, *optional*):
|
142 |
+
One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
|
143 |
+
to make generation deterministic.
|
144 |
+
num_inference_steps (`int`, *optional*, defaults to 50):
|
145 |
+
The number of denoising steps. More denoising steps usually lead to a higher quality image at the
|
146 |
+
expense of slower inference.
|
147 |
+
output_type (`str`, *optional*, defaults to `"pil"`):
|
148 |
+
The output format of the generate image. Choose between
|
149 |
+
[PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
|
150 |
+
return_dict (`bool`, *optional*, defaults to `True`):
|
151 |
+
Whether or not to return a [`~pipelines.ImagePipelineOutput`] instead of a plain tuple.
|
152 |
+
|
153 |
+
Returns:
|
154 |
+
[`~pipelines.ImagePipelineOutput`] or `tuple`: [`~pipelines.model.ImagePipelineOutput`] if `return_dict` is
|
155 |
+
True, otherwise a `tuple. When returning a tuple, the first element is a list with the generated images.
|
156 |
+
"""
|
157 |
+
# self.unet.config.sample_size = (64, 64) # (135,180)
|
158 |
+
# self.unet.config.sample_size = (135,180)
|
159 |
+
if self.resolution_type == "crack":
|
160 |
+
self.unet.config.sample_size = (64,64)
|
161 |
+
elif self.resolution_type == "crack_256":
|
162 |
+
self.unet.config.sample_size = (256,256)
|
163 |
+
else:
|
164 |
+
sc = 1080 // self.resolution
|
165 |
+
latent_size = (self.resolution // 4, 1440 // (sc*4))
|
166 |
+
self.unet.config.sample_size = latent_size
|
167 |
+
#
|
168 |
+
if not isinstance(self.unet.config.sample_size, tuple):
|
169 |
+
self.unet.config.sample_size = (self.unet.config.sample_size, self.unet.config.sample_size)
|
170 |
+
|
171 |
+
if segmap is None:
|
172 |
+
print("Didn't inpute any segmap, use the empty as the input")
|
173 |
+
segmap = torch.zeros(batch_size,self.unet.config.segmap_channels, self.unet.config.sample_size[0], self.unet.config.sample_size[1])
|
174 |
+
segmap = segmap.to(self.device).type(self.torch_dtype)
|
175 |
+
latents = randn_tensor(
|
176 |
+
(batch_size, self.unet.config.in_channels, self.unet.config.sample_size[0], self.unet.config.sample_size[1]),
|
177 |
+
generator=generator,
|
178 |
+
)
|
179 |
+
latents = latents.to(self.device).type(self.torch_dtype)
|
180 |
+
|
181 |
+
# scale the initial noise by the standard deviation required by the scheduler (need to check)
|
182 |
+
latents = latents * self.scheduler.init_noise_sigma
|
183 |
+
|
184 |
+
self.scheduler.set_timesteps(num_inference_steps=num_inference_steps)
|
185 |
+
|
186 |
+
# prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
|
187 |
+
accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
|
188 |
+
|
189 |
+
extra_kwargs = {}
|
190 |
+
if accepts_eta:
|
191 |
+
extra_kwargs["eta"] = eta
|
192 |
+
|
193 |
+
step_latent = []
|
194 |
+
learn_sigma = True if hasattr(self.scheduler, "variance_type") else False
|
195 |
+
for i, t in enumerate(self.progress_bar(self.scheduler.timesteps)):
|
196 |
+
|
197 |
+
latent_model_input = self.scheduler.scale_model_input(latents, t)
|
198 |
+
# predict the noise residual
|
199 |
+
noise_prediction = self.unet(latent_model_input, segmap, t).sample
|
200 |
+
# compute the previous noisy sample x_t -> x_t-1
|
201 |
+
|
202 |
+
|
203 |
+
if learn_sigma and "learn" in self.scheduler.variance_type:
|
204 |
+
model_pred, var_pred = torch.split(noise_prediction, latents.shape[1], dim=1)
|
205 |
+
else:
|
206 |
+
model_pred = noise_prediction
|
207 |
+
if s > 1.0:
|
208 |
+
model_output_zero = self.unet(latent_model_input, torch.zeros_like(segmap), t).sample
|
209 |
+
if learn_sigma and "learn" in self.scheduler.variance_type:
|
210 |
+
model_output_zero,_ = torch.split(model_output_zero, latents.shape[1], dim=1)
|
211 |
+
model_pred = model_pred + s * (model_pred - model_output_zero)
|
212 |
+
if learn_sigma and "learn" in self.scheduler.variance_type:
|
213 |
+
recombined = torch.cat((model_pred, var_pred), dim=1)
|
214 |
+
# when apply different scheduler, mean only !!
|
215 |
+
if learn_sigma and "learn" in self.scheduler.variance_type:
|
216 |
+
latents = self.scheduler.step(recombined, t, latents, **extra_kwargs).prev_sample
|
217 |
+
else:
|
218 |
+
latents = self.scheduler.step(noise_prediction, t, latents, **extra_kwargs).prev_sample
|
219 |
+
|
220 |
+
if every_step_save is not None:
|
221 |
+
if (i+1) % every_step_save == 0:
|
222 |
+
step_latent.append(copy.deepcopy(latents))
|
223 |
+
|
224 |
+
# decode the image latents with the VAE
|
225 |
+
if every_step_save is not None:
|
226 |
+
image = []
|
227 |
+
for i, l in enumerate(step_latent):
|
228 |
+
l /= self.vae.config.scaling_factor # (0.18215)
|
229 |
+
#latents /= 7.706491063029163
|
230 |
+
l = self.vae.decode(l, segmap)
|
231 |
+
l = (l / 2 + 0.5).clamp(0, 1)
|
232 |
+
l = l.cpu().permute(0, 2, 3, 1).numpy()
|
233 |
+
if output_type == "pil":
|
234 |
+
l = self.numpy_to_pil(l)
|
235 |
+
image.append(l)
|
236 |
+
else:
|
237 |
+
latents /= self.vae.config.scaling_factor#(0.18215)
|
238 |
+
#latents /= 7.706491063029163
|
239 |
+
# image = self.vae.decode(latents, segmap).sample
|
240 |
+
image = self.vae.decode(latents, return_dict=False)[0]
|
241 |
+
image = (image / 2 + 0.5).clamp(0, 1)
|
242 |
+
image = image.cpu().permute(0, 2, 3, 1).numpy()
|
243 |
+
if output_type == "pil":
|
244 |
+
image = self.numpy_to_pil(image)
|
245 |
+
|
246 |
+
if not return_dict:
|
247 |
+
return (image,)
|
248 |
+
|
249 |
+
return ImagePipelineOutput(images=image)
|
250 |
+
|
251 |
+
|
252 |
+
class SDMPipeline(DiffusionPipeline):
|
253 |
+
r"""
|
254 |
+
This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the
|
255 |
+
library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)
|
256 |
+
|
257 |
+
Parameters:
|
258 |
+
vae ([`VQModel`]):
|
259 |
+
Vector-quantized (VQ) Model to encode and decode images to and from latent representations.
|
260 |
+
unet ([`UNet2DModel`]): U-Net architecture to denoise the encoded image latents.
|
261 |
+
scheduler ([`SchedulerMixin`]):
|
262 |
+
[`DDIMScheduler`] is to be used in combination with `unet` to denoise the encoded image latents.
|
263 |
+
"""
|
264 |
+
|
265 |
+
def __init__(self, unet: UNet2DModel, scheduler: DDIMScheduler, torch_dtype=torch.float16, vae=None):
|
266 |
+
super().__init__()
|
267 |
+
self.register_modules(unet=unet, scheduler=scheduler)
|
268 |
+
self.torch_dtype = torch_dtype
|
269 |
+
|
270 |
+
@torch.no_grad()
|
271 |
+
def __call__(
|
272 |
+
self,
|
273 |
+
segmap = None,
|
274 |
+
batch_size: int = 8,
|
275 |
+
generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
|
276 |
+
eta: float = 0.0,
|
277 |
+
num_inference_steps: int = 1000,
|
278 |
+
output_type: Optional[str] = "pil",
|
279 |
+
return_dict: bool = True,
|
280 |
+
s: int = 1,
|
281 |
+
**kwargs,
|
282 |
+
) -> Union[Tuple, ImagePipelineOutput]:
|
283 |
+
r"""
|
284 |
+
Args:
|
285 |
+
batch_size (`int`, *optional*, defaults to 1):
|
286 |
+
Number of images to generate.
|
287 |
+
generator (`torch.Generator`, *optional*):
|
288 |
+
One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
|
289 |
+
to make generation deterministic.
|
290 |
+
num_inference_steps (`int`, *optional*, defaults to 50):
|
291 |
+
The number of denoising steps. More denoising steps usually lead to a higher quality image at the
|
292 |
+
expense of slower inference.
|
293 |
+
output_type (`str`, *optional*, defaults to `"pil"`):
|
294 |
+
The output format of the generate image. Choose between
|
295 |
+
[PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
|
296 |
+
return_dict (`bool`, *optional*, defaults to `True`):
|
297 |
+
Whether or not to return a [`~pipelines.ImagePipelineOutput`] instead of a plain tuple.
|
298 |
+
|
299 |
+
Returns:
|
300 |
+
[`~pipelines.ImagePipelineOutput`] or `tuple`: [`~pipelines.model.ImagePipelineOutput`] if `return_dict` is
|
301 |
+
True, otherwise a `tuple. When returning a tuple, the first element is a list with the generated images.
|
302 |
+
"""
|
303 |
+
self.unet.config.sample_size = (270,360)
|
304 |
+
if not isinstance(self.unet.config.sample_size, tuple):
|
305 |
+
self.unet.config.sample_size = (self.unet.config.sample_size, self.unet.config.sample_size)
|
306 |
+
|
307 |
+
if segmap is None:
|
308 |
+
print("Didn't inpute any segmap, use the empty as the input")
|
309 |
+
segmap = torch.zeros(batch_size,self.unet.config.segmap_channels, self.unet.config.sample_size[0], self.unet.config.sample_size[1])
|
310 |
+
segmap = segmap.to(self.device).type(self.torch_dtype)
|
311 |
+
latents = randn_tensor(
|
312 |
+
(batch_size, self.unet.config.in_channels, self.unet.config.sample_size[0], self.unet.config.sample_size[1]),
|
313 |
+
generator=generator,
|
314 |
+
)
|
315 |
+
|
316 |
+
latents = latents.to(self.device).type(self.torch_dtype)
|
317 |
+
|
318 |
+
# scale the initial noise by the standard deviation required by the scheduler (need to check)
|
319 |
+
latents = latents * self.scheduler.init_noise_sigma
|
320 |
+
|
321 |
+
self.scheduler.set_timesteps(num_inference_steps)
|
322 |
+
|
323 |
+
# prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
|
324 |
+
accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
|
325 |
+
|
326 |
+
extra_kwargs = {}
|
327 |
+
if accepts_eta:
|
328 |
+
extra_kwargs["eta"] = eta
|
329 |
+
|
330 |
+
for t in self.progress_bar(self.scheduler.timesteps):
|
331 |
+
latent_model_input = self.scheduler.scale_model_input(latents, t)
|
332 |
+
# predict the noise residual
|
333 |
+
noise_prediction = self.unet(latent_model_input, segmap, t).sample
|
334 |
+
|
335 |
+
#noise_prediction = noise_prediction[]
|
336 |
+
|
337 |
+
if s > 1.0:
|
338 |
+
model_output_zero = self.unet(latent_model_input, torch.zeros_like(segmap), t).sample
|
339 |
+
noise_prediction[:, :3] = model_output_zero[:, :3] + s * (noise_prediction[:, :3] - model_output_zero[:, :3])
|
340 |
+
|
341 |
+
#noise_prediction = noise_prediction[:, :3]
|
342 |
+
|
343 |
+
# compute the previous noisy sample x_t -> x_t-1
|
344 |
+
#breakpoint()
|
345 |
+
latents = self.scheduler.step(noise_prediction, t, latents, **extra_kwargs).prev_sample
|
346 |
+
|
347 |
+
# decode the image latents with the VAE
|
348 |
+
# latents /= self.vae.config.scaling_factor#(0.18215)
|
349 |
+
# image = self.vae.decode(latents).sample
|
350 |
+
image = latents
|
351 |
+
#image = (image + 1) / 2.0
|
352 |
+
image = (image / 2 + 0.5).clamp(0, 1)
|
353 |
+
image = image.cpu().permute(0, 2, 3, 1).numpy()
|
354 |
+
if output_type == "pil":
|
355 |
+
image = self.numpy_to_pil(image)
|
356 |
+
|
357 |
+
if not return_dict:
|
358 |
+
return (image,)
|
359 |
+
|
360 |
+
return ImagePipelineOutput(images=image)
|
361 |
+
|
diffusion_module/utils/__pycache__/LSDMPipeline_expandDataset.cpython-39.pyc
ADDED
Binary file (5.82 kB). View file
|
|
diffusion_module/utils/__pycache__/Pipline.cpython-310.pyc
ADDED
Binary file (8.22 kB). View file
|
|
diffusion_module/utils/__pycache__/Pipline.cpython-39.pyc
ADDED
Binary file (8.52 kB). View file
|
|
diffusion_module/utils/__pycache__/loss.cpython-39.pyc
ADDED
Binary file (4.06 kB). View file
|
|
diffusion_module/utils/loss.py
ADDED
@@ -0,0 +1,149 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""
|
2 |
+
Helpers for various likelihood-based losses. These are ported from the original
|
3 |
+
Ho et al. diffusion models codebase:
|
4 |
+
https://github.com/hojonathanho/diffusion/blob/1e0dceb3b3495bbe19116a5e1b3596cd0706c543/diffusion_tf/utils.py
|
5 |
+
"""
|
6 |
+
|
7 |
+
import numpy as np
|
8 |
+
|
9 |
+
import torch as th
|
10 |
+
|
11 |
+
|
12 |
+
def normal_kl(mean1, logvar1, mean2, logvar2):
|
13 |
+
"""
|
14 |
+
Compute the KL divergence between two gaussians.
|
15 |
+
|
16 |
+
Shapes are automatically broadcasted, so batches can be compared to
|
17 |
+
scalars, among other use cases.
|
18 |
+
"""
|
19 |
+
tensor = None
|
20 |
+
for obj in (mean1, logvar1, mean2, logvar2):
|
21 |
+
if isinstance(obj, th.Tensor):
|
22 |
+
tensor = obj
|
23 |
+
break
|
24 |
+
assert tensor is not None, "at least one argument must be a Tensor"
|
25 |
+
|
26 |
+
# Force variances to be Tensors. Broadcasting helps convert scalars to
|
27 |
+
# Tensors, but it does not work for th.exp().
|
28 |
+
logvar1, logvar2 = [
|
29 |
+
x if isinstance(x, th.Tensor) else th.tensor(x).to(tensor)
|
30 |
+
for x in (logvar1, logvar2)
|
31 |
+
]
|
32 |
+
|
33 |
+
return 0.5 * (
|
34 |
+
-1.0
|
35 |
+
+ logvar2
|
36 |
+
- logvar1
|
37 |
+
+ th.exp(logvar1 - logvar2)
|
38 |
+
+ ((mean1 - mean2) ** 2) * th.exp(-logvar2)
|
39 |
+
)
|
40 |
+
|
41 |
+
|
42 |
+
def approx_standard_normal_cdf(x):
|
43 |
+
"""
|
44 |
+
A fast approximation of the cumulative distribution function of the
|
45 |
+
standard normal.
|
46 |
+
"""
|
47 |
+
return 0.5 * (1.0 + th.tanh(np.sqrt(2.0 / np.pi) * (x + 0.044715 * th.pow(x, 3))))
|
48 |
+
|
49 |
+
|
50 |
+
def discretized_gaussian_log_likelihood(x, *, means, log_scales):
|
51 |
+
"""
|
52 |
+
Compute the log-likelihood of a Gaussian distribution discretizing to a
|
53 |
+
given image.
|
54 |
+
|
55 |
+
:param x: the target images. It is assumed that this was uint8 values,
|
56 |
+
rescaled to the range [-1, 1].
|
57 |
+
:param means: the Gaussian mean Tensor.
|
58 |
+
:param log_scales: the Gaussian log stddev Tensor.
|
59 |
+
:return: a tensor like x of log probabilities (in nats).
|
60 |
+
"""
|
61 |
+
assert x.shape == means.shape == log_scales.shape
|
62 |
+
centered_x = x - means
|
63 |
+
inv_stdv = th.exp(-log_scales)
|
64 |
+
plus_in = inv_stdv * (centered_x + 1.0 / 255.0)
|
65 |
+
cdf_plus = approx_standard_normal_cdf(plus_in)
|
66 |
+
min_in = inv_stdv * (centered_x - 1.0 / 255.0)
|
67 |
+
cdf_min = approx_standard_normal_cdf(min_in)
|
68 |
+
log_cdf_plus = th.log(cdf_plus.clamp(min=1e-12))
|
69 |
+
log_one_minus_cdf_min = th.log((1.0 - cdf_min).clamp(min=1e-12))
|
70 |
+
cdf_delta = cdf_plus - cdf_min
|
71 |
+
log_probs = th.where(
|
72 |
+
x < -0.999,
|
73 |
+
log_cdf_plus,
|
74 |
+
th.where(x > 0.999, log_one_minus_cdf_min, th.log(cdf_delta.clamp(min=1e-12))),
|
75 |
+
)
|
76 |
+
assert log_probs.shape == x.shape
|
77 |
+
return log_probs
|
78 |
+
|
79 |
+
def variance_KL_loss(latents, noisy_latents, timesteps, model_pred_mean, model_pred_var, noise_scheduler,posterior_mean_coef1, posterior_mean_coef2, posterior_log_variance_clipped):
|
80 |
+
model_pred_mean = model_pred_mean.detach()
|
81 |
+
true_mean = (
|
82 |
+
posterior_mean_coef1.to(device=timesteps.device)[timesteps].float()[..., None, None, None] * latents
|
83 |
+
+ posterior_mean_coef2.to(device=timesteps.device)[timesteps].float()[..., None, None, None] * noisy_latents
|
84 |
+
)
|
85 |
+
|
86 |
+
true_log_variance_clipped = posterior_log_variance_clipped.to(device=timesteps.device)[timesteps].float()[
|
87 |
+
..., None, None, None]
|
88 |
+
|
89 |
+
if noise_scheduler.variance_type == "learned":
|
90 |
+
model_log_variance = model_pred_var
|
91 |
+
#model_pred_var = th.exp(model_log_variance)
|
92 |
+
else:
|
93 |
+
min_log = true_log_variance_clipped
|
94 |
+
max_log = th.log(noise_scheduler.betas.to(device=timesteps.device)[timesteps].float()[..., None, None, None])
|
95 |
+
frac = (model_pred_var + 1) / 2
|
96 |
+
model_log_variance = frac * max_log + (1 - frac) * min_log
|
97 |
+
#model_pred_var = th.exp(model_log_variance)
|
98 |
+
|
99 |
+
sqrt_recip_alphas_cumprod = th.sqrt(1.0 / noise_scheduler.alphas_cumprod)
|
100 |
+
sqrt_recipm1_alphas_cumprod = th.sqrt(1.0 / noise_scheduler.alphas_cumprod - 1)
|
101 |
+
|
102 |
+
pred_xstart = (sqrt_recip_alphas_cumprod.to(device=timesteps.device)[timesteps].float()[
|
103 |
+
..., None, None, None] * noisy_latents
|
104 |
+
- sqrt_recipm1_alphas_cumprod.to(device=timesteps.device)[timesteps].float()[
|
105 |
+
..., None, None, None] * model_pred_mean)
|
106 |
+
|
107 |
+
model_mean = (
|
108 |
+
posterior_mean_coef1.to(device=timesteps.device)[timesteps].float()[..., None, None, None] * pred_xstart
|
109 |
+
+ posterior_mean_coef2.to(device=timesteps.device)[timesteps].float()[..., None, None, None] * noisy_latents
|
110 |
+
)
|
111 |
+
|
112 |
+
# model_mean = out["mean"] model_log_variance = out["log_variance"]
|
113 |
+
kl = normal_kl(
|
114 |
+
true_mean, true_log_variance_clipped, model_mean, model_log_variance
|
115 |
+
)
|
116 |
+
kl = kl.mean() / np.log(2.0)
|
117 |
+
|
118 |
+
decoder_nll = -discretized_gaussian_log_likelihood(
|
119 |
+
latents, means=model_mean, log_scales=0.5 * model_log_variance
|
120 |
+
)
|
121 |
+
assert decoder_nll.shape == latents.shape
|
122 |
+
decoder_nll = decoder_nll.mean() / np.log(2.0)
|
123 |
+
|
124 |
+
# At the first timestep return the decoder NLL,
|
125 |
+
# otherwise return KL(q(x_{t-1}|x_t,x_0) || p(x_{t-1}|x_t))
|
126 |
+
kl_loss = th.where((timesteps == 0), decoder_nll, kl).mean()
|
127 |
+
return kl_loss
|
128 |
+
|
129 |
+
def get_variance(noise_scheduler):
|
130 |
+
alphas_cumprod_prev = th.cat([th.tensor([1.0]), noise_scheduler.alphas_cumprod[:-1]])
|
131 |
+
|
132 |
+
posterior_mean_coef1 = (
|
133 |
+
noise_scheduler.betas * th.sqrt(alphas_cumprod_prev) / (1.0 - noise_scheduler.alphas_cumprod)
|
134 |
+
)
|
135 |
+
|
136 |
+
posterior_mean_coef2 = (
|
137 |
+
(1.0 - alphas_cumprod_prev)
|
138 |
+
* th.sqrt(noise_scheduler.alphas)
|
139 |
+
/ (1.0 - noise_scheduler.alphas_cumprod)
|
140 |
+
)
|
141 |
+
|
142 |
+
posterior_variance = (
|
143 |
+
noise_scheduler.betas * (1.0 - alphas_cumprod_prev) / (1.0 - noise_scheduler.alphas_cumprod)
|
144 |
+
)
|
145 |
+
posterior_log_variance_clipped = th.log(
|
146 |
+
th.cat([posterior_variance[1][..., None], posterior_variance[1:]])
|
147 |
+
)
|
148 |
+
#res = posterior_log_variance_clipped.to(device=timesteps.device)[timesteps].float()
|
149 |
+
return posterior_mean_coef1, posterior_mean_coef2, posterior_log_variance_clipped #res[..., None, None, None]
|
diffusion_module/utils/noise_sampler.py
ADDED
@@ -0,0 +1,16 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from sklearn.mixture import GaussianMixture
|
2 |
+
|
3 |
+
def get_noise_sampler(sample_type='gau'):
|
4 |
+
if sample_type == 'gau':
|
5 |
+
sampler = lambda latnt_sz: torch.randn_like(latnt_sz)
|
6 |
+
elif sample_type == 'gau_offset':
|
7 |
+
sampler = lambda latnt_sz: torch.randn_like(latnt_sz) + (torch.randn_like(latnt_sz))
|
8 |
+
...
|
9 |
+
elif sample_type == 'gmm':
|
10 |
+
...
|
11 |
+
else:
|
12 |
+
...
|
13 |
+
return
|
14 |
+
|
15 |
+
if __name__ == "__main__":
|
16 |
+
...
|
diffusion_module/utils/scheduler_factory.py
ADDED
@@ -0,0 +1,300 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from functools import partial
|
2 |
+
from diffusers import DDPMScheduler, DPMSolverMultistepScheduler, UniPCMultistepScheduler, DPMSolverSinglestepScheduler
|
3 |
+
from diffusers.pipeline_utils import DiffusionPipeline
|
4 |
+
import torch
|
5 |
+
from diffusers.configuration_utils import ConfigMixin, register_to_config
|
6 |
+
from typing import List, Optional, Tuple, Union
|
7 |
+
import numpy as np
|
8 |
+
from diffusers.schedulers.scheduling_utils import SchedulerOutput
|
9 |
+
from diffusers.schedulers.scheduling_ddpm import DDPMSchedulerOutput
|
10 |
+
from diffusers.utils import randn_tensor, BaseOutput
|
11 |
+
|
12 |
+
|
13 |
+
### Testing the DDPM Scheduler for Variant
|
14 |
+
class ModifiedDDPMScheduler(DDPMScheduler):
|
15 |
+
def __init__(self, *args, **kwargs):
|
16 |
+
super().__init__(*args, **kwargs)
|
17 |
+
|
18 |
+
def step(
|
19 |
+
self,
|
20 |
+
model_output: torch.FloatTensor,
|
21 |
+
timestep: int,
|
22 |
+
sample: torch.FloatTensor,
|
23 |
+
generator=None,
|
24 |
+
return_dict: bool = True,
|
25 |
+
) -> Union[DDPMSchedulerOutput, Tuple]:
|
26 |
+
"""
|
27 |
+
Predict the sample at the previous timestep by reversing the SDE. Core function to propagate the diffusion
|
28 |
+
process from the learned model outputs (most often the predicted noise).
|
29 |
+
|
30 |
+
Args:
|
31 |
+
model_output (`torch.FloatTensor`): direct output from learned diffusion model.
|
32 |
+
timestep (`int`): current discrete timestep in the diffusion chain.
|
33 |
+
sample (`torch.FloatTensor`):
|
34 |
+
current instance of sample being created by diffusion process.
|
35 |
+
generator: random number generator.
|
36 |
+
return_dict (`bool`): option for returning tuple rather than DDPMSchedulerOutput class
|
37 |
+
|
38 |
+
Returns:
|
39 |
+
[`~schedulers.scheduling_utils.DDPMSchedulerOutput`] or `tuple`:
|
40 |
+
[`~schedulers.scheduling_utils.DDPMSchedulerOutput`] if `return_dict` is True, otherwise a `tuple`. When
|
41 |
+
returning a tuple, the first element is the sample tensor.
|
42 |
+
|
43 |
+
"""
|
44 |
+
t = timestep
|
45 |
+
|
46 |
+
prev_t = self.previous_timestep(t)
|
47 |
+
|
48 |
+
if model_output.shape[1] == sample.shape[1] * 2 and self.variance_type in ["learned", "learned_range"]:
|
49 |
+
print("Conidtion is trigger")
|
50 |
+
|
51 |
+
model_output, predicted_variance = torch.split(model_output, sample.shape[1], dim=1)
|
52 |
+
# [2,3, 64, 128]
|
53 |
+
else:
|
54 |
+
predicted_variance = None
|
55 |
+
|
56 |
+
# 1. compute alphas, betas
|
57 |
+
alpha_prod_t = self.alphas_cumprod[t]
|
58 |
+
alpha_prod_t_prev = self.alphas_cumprod[prev_t] if prev_t >= 0 else self.one
|
59 |
+
beta_prod_t = 1 - alpha_prod_t
|
60 |
+
beta_prod_t_prev = 1 - alpha_prod_t_prev
|
61 |
+
current_alpha_t = alpha_prod_t / alpha_prod_t_prev
|
62 |
+
current_beta_t = 1 - current_alpha_t
|
63 |
+
|
64 |
+
# 2. compute predicted original sample from predicted noise also called
|
65 |
+
# "predicted x_0" of formula (15) from https://arxiv.org/pdf/2006.11239.pdf
|
66 |
+
if self.config.prediction_type == "epsilon":
|
67 |
+
pred_original_sample = (sample - beta_prod_t ** (0.5) * model_output) / alpha_prod_t ** (0.5)
|
68 |
+
|
69 |
+
elif self.config.prediction_type == "sample":
|
70 |
+
pred_original_sample = model_output
|
71 |
+
elif self.config.prediction_type == "v_prediction":
|
72 |
+
pred_original_sample = (alpha_prod_t**0.5) * sample - (beta_prod_t**0.5) * model_output
|
73 |
+
else:
|
74 |
+
raise ValueError(
|
75 |
+
f"prediction_type given as {self.config.prediction_type} must be one of `epsilon`, `sample` or"
|
76 |
+
" `v_prediction` for the DDPMScheduler."
|
77 |
+
)
|
78 |
+
|
79 |
+
# 3. Clip or threshold "predicted x_0"
|
80 |
+
if self.config.thresholding:
|
81 |
+
pred_original_sample = self._threshold_sample(pred_original_sample)
|
82 |
+
elif self.config.clip_sample:
|
83 |
+
pred_original_sample = pred_original_sample.clamp(
|
84 |
+
-self.config.clip_sample_range, self.config.clip_sample_range
|
85 |
+
)
|
86 |
+
|
87 |
+
# 4. Compute coefficients for pred_original_sample x_0 and current sample x_t
|
88 |
+
# See formula (7) from https://arxiv.org/pdf/2006.11239.pdf
|
89 |
+
pred_original_sample_coeff = (alpha_prod_t_prev ** (0.5) * current_beta_t) / beta_prod_t
|
90 |
+
current_sample_coeff = current_alpha_t ** (0.5) * beta_prod_t_prev / beta_prod_t
|
91 |
+
|
92 |
+
# 5. Compute predicted previous sample µ_t
|
93 |
+
# See formula (7) from https://arxiv.org/pdf/2006.11239.pdf
|
94 |
+
pred_prev_sample = pred_original_sample_coeff * pred_original_sample + current_sample_coeff * sample
|
95 |
+
|
96 |
+
# 6. Add noise
|
97 |
+
variance = 0
|
98 |
+
if t > 0:
|
99 |
+
device = model_output.device
|
100 |
+
variance_noise = randn_tensor(
|
101 |
+
model_output.shape, generator=generator, device=device, dtype=model_output.dtype
|
102 |
+
)
|
103 |
+
if self.variance_type == "fixed_small_log":
|
104 |
+
variance = self._get_variance(t, predicted_variance=predicted_variance) * variance_noise
|
105 |
+
|
106 |
+
elif self.variance_type == "learned_range":
|
107 |
+
variance = self._get_variance(t, predicted_variance=predicted_variance)
|
108 |
+
variance = torch.exp(0.5 * variance) * variance_noise
|
109 |
+
|
110 |
+
else:
|
111 |
+
variance = (self._get_variance(t, predicted_variance=predicted_variance) ** 0.5) * variance_noise
|
112 |
+
|
113 |
+
pred_prev_sample = pred_prev_sample + variance
|
114 |
+
print(pred_prev_sample.shape)
|
115 |
+
if not return_dict:
|
116 |
+
return (pred_prev_sample,)
|
117 |
+
|
118 |
+
return DDPMSchedulerOutput(prev_sample=pred_prev_sample, pred_original_sample=pred_original_sample)
|
119 |
+
|
120 |
+
|
121 |
+
class ModifiedUniPCScheduler(UniPCMultistepScheduler):
|
122 |
+
'''
|
123 |
+
This is the modification of UniPCMultistepScheduler, which is the same as UniPCMultistepScheduler except for the _get_variance function.
|
124 |
+
'''
|
125 |
+
def __init__(self, variance_type: str = "fixed_small", *args, **kwargs):
|
126 |
+
super().__init__(*args, **kwargs)
|
127 |
+
self.custom_timesteps = False
|
128 |
+
self.variance_type=variance_type
|
129 |
+
self.config.timestep_spacing="leading"
|
130 |
+
def previous_timestep(self, timestep):
|
131 |
+
if self.custom_timesteps:
|
132 |
+
index = (self.timesteps == timestep).nonzero(as_tuple=True)[0][0]
|
133 |
+
if index == self.timesteps.shape[0] - 1:
|
134 |
+
prev_t = torch.tensor(-1)
|
135 |
+
else:
|
136 |
+
prev_t = self.timesteps[index + 1]
|
137 |
+
else:
|
138 |
+
num_inference_steps = (
|
139 |
+
self.num_inference_steps if self.num_inference_steps else self.config.num_train_timesteps
|
140 |
+
)
|
141 |
+
prev_t = timestep - self.config.num_train_timesteps // num_inference_steps
|
142 |
+
|
143 |
+
return prev_t
|
144 |
+
|
145 |
+
def _get_variance(self, t, predicted_variance=None, variance_type="learned_range"):
|
146 |
+
prev_t = self.previous_timestep(t)
|
147 |
+
|
148 |
+
alpha_prod_t = self.alphas_cumprod[t]
|
149 |
+
alpha_prod_t_prev = self.alphas_cumprod[prev_t] if prev_t >= 0 else self.one
|
150 |
+
current_beta_t = 1 - alpha_prod_t / alpha_prod_t_prev
|
151 |
+
|
152 |
+
variance = (1 - alpha_prod_t_prev) / (1 - alpha_prod_t) * current_beta_t
|
153 |
+
|
154 |
+
variance = torch.clamp(variance, min=1e-20)
|
155 |
+
|
156 |
+
if variance_type is None:
|
157 |
+
variance_type = self.config.variance_type
|
158 |
+
|
159 |
+
if variance_type == "fixed_small":
|
160 |
+
variance = variance
|
161 |
+
elif variance_type == "fixed_small_log":
|
162 |
+
variance = torch.log(variance)
|
163 |
+
variance = torch.exp(0.5 * variance)
|
164 |
+
elif variance_type == "fixed_large":
|
165 |
+
variance = current_beta_t
|
166 |
+
elif variance_type == "fixed_large_log":
|
167 |
+
variance = torch.log(current_beta_t)
|
168 |
+
elif variance_type == "learned":
|
169 |
+
return predicted_variance
|
170 |
+
elif variance_type == "learned_range":
|
171 |
+
min_log = torch.log(variance)
|
172 |
+
max_log = torch.log(current_beta_t)
|
173 |
+
frac = (predicted_variance + 1) / 2
|
174 |
+
variance = frac * max_log + (1 - frac) * min_log
|
175 |
+
|
176 |
+
return variance
|
177 |
+
|
178 |
+
def step(self, model_output: torch.FloatTensor, timestep: int, sample: torch.FloatTensor, return_dict: bool = True) -> Union[SchedulerOutput, Tuple]:
|
179 |
+
|
180 |
+
if model_output.shape[1] == sample.shape[1] * 2 and self.variance_type in ["learned", "learned_range"]:
|
181 |
+
print("condition using predicted_variance is trigger")
|
182 |
+
model_output, predicted_variance = torch.split(model_output, sample.shape[1], dim=1)
|
183 |
+
else:
|
184 |
+
predicted_variance = None
|
185 |
+
|
186 |
+
super_output = super().step(model_output, timestep, sample, return_dict=False)
|
187 |
+
prev_sample = super_output[0]
|
188 |
+
# breakpoint()
|
189 |
+
variance = 0
|
190 |
+
if timestep > 0:
|
191 |
+
device = model_output.device
|
192 |
+
variance_noise = randn_tensor(
|
193 |
+
model_output.shape, generator=None, device=device, dtype=model_output.dtype
|
194 |
+
)
|
195 |
+
if self.variance_type == "fixed_small_log":
|
196 |
+
variance = self._get_variance(timestep, predicted_variance=predicted_variance) * variance_noise
|
197 |
+
elif self.variance_type == "learned_range":
|
198 |
+
# breakpoint()
|
199 |
+
variance = self._get_variance(timestep, predicted_variance=predicted_variance)
|
200 |
+
variance = torch.exp(0.5 * variance) * variance_noise
|
201 |
+
# breakpoint()
|
202 |
+
else:
|
203 |
+
variance = (self._get_variance(timestep, predicted_variance=predicted_variance) ** 0.5) * variance_noise
|
204 |
+
|
205 |
+
|
206 |
+
# breakpoint()
|
207 |
+
print("time step is ", timestep)
|
208 |
+
prev_sample = prev_sample + variance
|
209 |
+
|
210 |
+
if not return_dict:
|
211 |
+
return (prev_sample,)
|
212 |
+
|
213 |
+
return DDPMSchedulerOutput(prev_sample=prev_sample,pred_original_sample=prev_sample)
|
214 |
+
|
215 |
+
#return SchedulerOutput(prev_sample=prev_sample)
|
216 |
+
|
217 |
+
|
218 |
+
def build_proc(sch_cfg=None, _sch=None, **kwargs):
|
219 |
+
if kwargs:
|
220 |
+
return _sch(**kwargs)
|
221 |
+
|
222 |
+
type_str = str(type(sch_cfg))
|
223 |
+
if 'dict' in type_str:
|
224 |
+
return _sch.from_config(**sch_cfg)
|
225 |
+
return _sch.from_config(sch_cfg, subfolder="scheduler")
|
226 |
+
|
227 |
+
scheduler_factory = {
|
228 |
+
'UniPC' : partial(build_proc, _sch=UniPCMultistepScheduler),
|
229 |
+
'modifiedUniPC' : partial(build_proc, _sch=ModifiedUniPCScheduler),
|
230 |
+
# DPM family
|
231 |
+
'DDPM' : partial(build_proc, _sch=DDPMScheduler),
|
232 |
+
'DPMSolver' : partial(build_proc, _sch=DPMSolverMultistepScheduler, algorithm_type='dpmsolver'),
|
233 |
+
'DPMSolver++' : partial(build_proc, _sch=DPMSolverMultistepScheduler),
|
234 |
+
'DPMSolverSingleStep' : partial(build_proc, _sch=DPMSolverSinglestepScheduler)
|
235 |
+
|
236 |
+
}
|
237 |
+
|
238 |
+
def scheduler_setup(pipe : DiffusionPipeline = None, scheduler_type : str = 'UniPC', from_config=None, **kwargs):
|
239 |
+
if not isinstance(pipe, DiffusionPipeline):
|
240 |
+
raise TypeError(f'pipe should be DiffusionPipeline, but given {type(pipe)}\n')
|
241 |
+
|
242 |
+
sch_cfg = from_config if from_config else pipe.scheduler.config
|
243 |
+
#sch_cfg = diffusers.configuration_utils.FrozenDict({**sch_cfg, 'solver_order':3})
|
244 |
+
#pipe.scheduler = scheduler_factory[scheduler_type](**kwargs) if kwargs \
|
245 |
+
# else scheduler_factory[scheduler_type](sch_cfg)
|
246 |
+
|
247 |
+
# pipe.scheduler = DPMSolverSinglestepScheduler()
|
248 |
+
# #pipe.scheduler = DDPMScheduler(beta_schedule="linear", variance_type="learned_range")
|
249 |
+
# print(pipe.scheduler)
|
250 |
+
print("Scheduler type in Scheduler_factory.py is Hard-coded to modifyUniPC, Please change it back to AutoDetect functionality if you want to change scheudler")
|
251 |
+
pipe.scheduler = ModifiedUniPCScheduler(variance_type="learned_range", )
|
252 |
+
# pipe.scheduler = ModifiedDDPMScheduler(beta_schedule="linear", variance_type="learned_range")
|
253 |
+
|
254 |
+
#pipe.scheduler = DDPMScheduler.from_pretrained("CompVis/stable-diffusion-v1-4", subfolder="scheduler")
|
255 |
+
#pipe.scheduler._get_variance = _get_variance
|
256 |
+
return pipe
|
257 |
+
|
258 |
+
# unittest of scheduler..
|
259 |
+
if __name__ == "__main__":
|
260 |
+
def ld_mod():
|
261 |
+
noise_scheduler = DDPMScheduler.from_pretrained("CompVis/stable-diffusion-v1-4", subfolder="scheduler")
|
262 |
+
vae = AutoencoderKL.from_pretrained("CompVis/stable-diffusion-v1-4", subfolder="vae").to("cuda").to(torch.float16)
|
263 |
+
unet = SDMUNet2DModel.from_pretrained("/data/harry/Data_generation/diffusers-main/examples/VAESDM/LDM-sdm-model/checkpoint-46000", subfolder="unet").to("cuda").to(torch.float16)
|
264 |
+
return noise_scheduler, vae, unet
|
265 |
+
|
266 |
+
from Pipline import SDMLDMPipeline
|
267 |
+
from diffusers import StableDiffusionPipeline
|
268 |
+
import torch
|
269 |
+
|
270 |
+
path = "CompVis/stable-diffusion-v1-4"
|
271 |
+
pipe = StableDiffusionPipeline.from_pretrained(path, torch_dtype=torch.float16)
|
272 |
+
|
273 |
+
# change scheduler
|
274 |
+
# customized args : once you customized, customize forever ~ no from_config
|
275 |
+
#pipe = scheduler_setup(pipe, 'DPMSolver++', thresholding=True)
|
276 |
+
# from_config
|
277 |
+
pipe = scheduler_setup(pipe, 'DPMSolverSingleStep')
|
278 |
+
|
279 |
+
pipe = pipe.to("cuda")
|
280 |
+
prompt = "a highly realistic photo of green turtle"
|
281 |
+
generator = torch.manual_seed(0)
|
282 |
+
# only 15 steps are needed for good results => 2-4 seconds on GPU
|
283 |
+
image = pipe(prompt, generator=generator, num_inference_steps=15).images[0]
|
284 |
+
# save image
|
285 |
+
image.save("turtle.png")
|
286 |
+
|
287 |
+
'''
|
288 |
+
# load & wrap submodules into pipe-API
|
289 |
+
noise_scheduler, vae, unet = ld_mod()
|
290 |
+
pipe = SDMLDMPipeline(
|
291 |
+
unet=unet,
|
292 |
+
vqvae=vae,
|
293 |
+
scheduler=noise_scheduler,
|
294 |
+
torch_dtype=torch.float16
|
295 |
+
)
|
296 |
+
|
297 |
+
# change scheduler
|
298 |
+
pipe = scheduler_setup(pipe, 'DPMSolverSingleStep')
|
299 |
+
pipe = pipe.to("cuda")
|
300 |
+
'''
|
evolution.py
ADDED
@@ -0,0 +1,102 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import random
|
2 |
+
import math
|
3 |
+
import numpy as np
|
4 |
+
from PIL import Image
|
5 |
+
from skimage.draw import line
|
6 |
+
from skimage import morphology
|
7 |
+
import cv2
|
8 |
+
|
9 |
+
def line_crosses_cracks(start, end, img):
|
10 |
+
rr, cc = line(start[0], start[1], end[0], end[1])
|
11 |
+
# Exclude the starting point from the line coordinates
|
12 |
+
if len(rr) > 1 and len(cc) > 1:
|
13 |
+
return np.any(img[rr[1:], cc[1:]] == 255)
|
14 |
+
return False
|
15 |
+
|
16 |
+
def random_walk(img_array, k=8, m=0.1, min_steps=50, max_steps=200, length=2, degree_range=30, seed=None):
|
17 |
+
|
18 |
+
if seed is not None:
|
19 |
+
random.seed(seed)
|
20 |
+
np.random.seed(seed)
|
21 |
+
|
22 |
+
|
23 |
+
img_array = cv2.ximgproc.thinning(img_array)
|
24 |
+
|
25 |
+
rows, cols = img_array.shape
|
26 |
+
# Find all white pixels (existing cracks)
|
27 |
+
white_pixels = np.column_stack(np.where(img_array == 255))
|
28 |
+
original_crack_count = len(white_pixels) # Count of original crack pixels
|
29 |
+
|
30 |
+
# Select k random starting points from the white pixels
|
31 |
+
if white_pixels.size == 0:
|
32 |
+
raise ValueError("No initial crack pixels found in the image.")
|
33 |
+
if k > len(white_pixels):
|
34 |
+
raise ValueError("k is greater than the number of existing crack pixels.")
|
35 |
+
initial_points = white_pixels[random.sample(range(len(white_pixels)), k)]
|
36 |
+
|
37 |
+
# Initialize step count for each initial point with a random value between min_steps and max_steps
|
38 |
+
step_counts = {i: random.randint(min_steps, max_steps) for i in range(k)}
|
39 |
+
# Initialize main direction for each initial point (0 to 360 degrees)
|
40 |
+
main_angles = {i: random.uniform(0, 360) for i in range(k)}
|
41 |
+
|
42 |
+
grown_crack_count = 0 # Count of newly grown crack pixels
|
43 |
+
|
44 |
+
# Start the random walk for each initial point
|
45 |
+
for idx, point in enumerate(initial_points):
|
46 |
+
current_pos = tuple(point)
|
47 |
+
current_steps = 0
|
48 |
+
while current_steps < step_counts[idx]:
|
49 |
+
# Check the crack ratio
|
50 |
+
current_ratio = np.sum(img_array == 255) / (rows * cols)
|
51 |
+
if current_ratio >= m:
|
52 |
+
return img_array, {'original_crack_count': original_crack_count, 'grown_crack_count': grown_crack_count}
|
53 |
+
|
54 |
+
# Generate a random direction within the fan-shaped area around the main angle
|
55 |
+
main_angle = main_angles[idx]
|
56 |
+
angle = math.radians(main_angle + random.uniform(-degree_range, degree_range))
|
57 |
+
|
58 |
+
# Determine the next position with the specified length
|
59 |
+
delta_row = length * math.sin(angle)
|
60 |
+
delta_col = length * math.cos(angle)
|
61 |
+
next_pos = (int(current_pos[0] + delta_row), int(current_pos[1] + delta_col))
|
62 |
+
|
63 |
+
# Check if the line from the current to the next position crosses existing cracks
|
64 |
+
if 0 <= next_pos[0] < rows and 0 <= next_pos[1] < cols and not line_crosses_cracks(current_pos, next_pos, img_array):
|
65 |
+
# Draw a line from the current position to the next position
|
66 |
+
rr, cc = line(current_pos[0], current_pos[1], next_pos[0], next_pos[1])
|
67 |
+
img_array[rr, cc] = 255 # Set the pixels along the line to white
|
68 |
+
grown_crack_count += len(rr) # Update the count of grown crack pixels
|
69 |
+
current_pos = next_pos
|
70 |
+
current_steps += 1
|
71 |
+
else:
|
72 |
+
# If the line crosses existing cracks or the next position is outside the boundaries, stop the walk for this point
|
73 |
+
break
|
74 |
+
|
75 |
+
return img_array, {'original_crack_count': original_crack_count, 'grown_crack_count': grown_crack_count}
|
76 |
+
|
77 |
+
# The rest of the test code remains the same.
|
78 |
+
# You can use this function in your test code to generate the image and get the counts.
|
79 |
+
|
80 |
+
|
81 |
+
# test code
|
82 |
+
if __name__ == "__main__":
|
83 |
+
# Updated parameters
|
84 |
+
k = 8 # Number of initial white pixels to start the random walk
|
85 |
+
m = 0.1 # Maximum ratio of crack pixels
|
86 |
+
min_steps = 50
|
87 |
+
max_steps = 200
|
88 |
+
img_path = '/data/leiqin/diffusion/huggingface_diffusers/crack_label_creator/random_walk/thindata_256/2.png'
|
89 |
+
img = Image.open(img_path)
|
90 |
+
img_array = np.array(img)
|
91 |
+
length = 2
|
92 |
+
|
93 |
+
# Perform the modified random walk
|
94 |
+
result_img_array_mod, pixels_dict = random_walk(img_array.copy(), k, m, min_steps, max_steps, length)
|
95 |
+
|
96 |
+
# Convert the result to an image
|
97 |
+
result_img_mod = Image.fromarray(result_img_array_mod.astype('uint8'))
|
98 |
+
|
99 |
+
# Save the resulting image
|
100 |
+
result_img_path_mod = 'resutls.png'
|
101 |
+
result_img_mod.save(result_img_path_mod)
|
102 |
+
print(pixels_dict)
|
figs/4.png
ADDED
figs/4_1.jpg
ADDED
figs/4_1.png
ADDED
figs/4_1_mask.png
ADDED
generate.py
ADDED
@@ -0,0 +1,106 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from diffusers.schedulers import UniPCMultistepScheduler
|
2 |
+
from diffusers import AutoencoderKL
|
3 |
+
from diffusion_module.unet import UNetModel
|
4 |
+
import torch
|
5 |
+
from diffusion_module.utils.LSDMPipeline_expandDataset import SDMLDMPipeline
|
6 |
+
from accelerate import Accelerator
|
7 |
+
from evolution import random_walk
|
8 |
+
import cv2
|
9 |
+
import numpy as np
|
10 |
+
|
11 |
+
def mask2onehot(data, num_classes):
|
12 |
+
# move to GPU and change data types
|
13 |
+
data = data.to(dtype=torch.int64)
|
14 |
+
|
15 |
+
# create one-hot label map
|
16 |
+
label_map = data
|
17 |
+
bs, _, h, w = label_map.size()
|
18 |
+
input_label = torch.FloatTensor(bs, num_classes, h, w).zero_().to(data.device)
|
19 |
+
input_semantics = input_label.scatter_(1, label_map, 1.0)
|
20 |
+
|
21 |
+
return input_semantics
|
22 |
+
|
23 |
+
def generate(img, pretrain_weight,seed=None):
|
24 |
+
|
25 |
+
noise_scheduler = UniPCMultistepScheduler()
|
26 |
+
vae = AutoencoderKL.from_pretrained("runwayml/stable-diffusion-v1-5", subfolder="vae")
|
27 |
+
latent_size = (64, 64)
|
28 |
+
unet = UNetModel(
|
29 |
+
image_size = latent_size,
|
30 |
+
in_channels=vae.config.latent_channels,
|
31 |
+
model_channels=256,
|
32 |
+
out_channels=vae.config.latent_channels,
|
33 |
+
num_res_blocks=2,
|
34 |
+
attention_resolutions=(2, 4, 8),
|
35 |
+
dropout=0,
|
36 |
+
channel_mult=(1, 2, 3, 4),
|
37 |
+
num_heads=8,
|
38 |
+
num_head_channels=-1,
|
39 |
+
num_heads_upsample=-1,
|
40 |
+
use_scale_shift_norm=True,
|
41 |
+
resblock_updown=True,
|
42 |
+
use_new_attention_order=False,
|
43 |
+
num_classes=151,
|
44 |
+
mask_emb="resize",
|
45 |
+
use_checkpoint=True,
|
46 |
+
SPADE_type="spade",
|
47 |
+
)
|
48 |
+
|
49 |
+
|
50 |
+
unet = unet.from_pretrained(pretrain_weight)
|
51 |
+
device = 'cpu'
|
52 |
+
if device != 'cpu':
|
53 |
+
mixed_precision = "fp16"
|
54 |
+
else:
|
55 |
+
mixed_precision = "no"
|
56 |
+
|
57 |
+
|
58 |
+
accelerator = Accelerator(
|
59 |
+
mixed_precision=mixed_precision,
|
60 |
+
cpu= True if device is 'cpu' else False
|
61 |
+
)
|
62 |
+
|
63 |
+
weight_dtype = torch.float32
|
64 |
+
if accelerator.mixed_precision == "fp16":
|
65 |
+
weight_dtype = torch.float16
|
66 |
+
|
67 |
+
unet,vae = accelerator.prepare(unet, vae)
|
68 |
+
vae.to(device=accelerator.device, dtype=weight_dtype)
|
69 |
+
pipeline = SDMLDMPipeline(
|
70 |
+
vae=accelerator.unwrap_model(vae),
|
71 |
+
unet=accelerator.unwrap_model(unet),
|
72 |
+
scheduler=noise_scheduler,
|
73 |
+
torch_dtype=weight_dtype,
|
74 |
+
resolution_type="crack"
|
75 |
+
)
|
76 |
+
"""
|
77 |
+
if accelerator.device != 'cpu':
|
78 |
+
pipeline.enable_xformers_memory_efficient_attention()
|
79 |
+
"""
|
80 |
+
pipeline = pipeline.to(accelerator.device)
|
81 |
+
pipeline.set_progress_bar_config(disable=False)
|
82 |
+
|
83 |
+
if seed is None:
|
84 |
+
generator = None
|
85 |
+
else:
|
86 |
+
generator = torch.Generator(device=accelerator.device).manual_seed(seed)
|
87 |
+
|
88 |
+
resized_s = cv2.resize(img, (64, 64), interpolation=cv2.INTER_AREA)
|
89 |
+
# 灰度图放大到255
|
90 |
+
_, binary_s = cv2.threshold(resized_s, 1, 255, cv2.THRESH_BINARY)
|
91 |
+
# 转换为0,1
|
92 |
+
tensor_s = torch.from_numpy(binary_s / 255)
|
93 |
+
# h,w -> 1,1,h,w
|
94 |
+
tensor_s = tensor_s.unsqueeze(0).unsqueeze(0)
|
95 |
+
onehot_skeletons=[]
|
96 |
+
onehot_s = mask2onehot(tensor_s, 151)
|
97 |
+
onehot_skeletons.append(onehot_s)
|
98 |
+
|
99 |
+
onehot_skeletons = torch.stack(onehot_skeletons, dim=1).squeeze(0)
|
100 |
+
onehot_skeletons = onehot_skeletons.to(dtype=weight_dtype,device=accelerator.device)
|
101 |
+
|
102 |
+
images = pipeline(onehot_skeletons, generator=generator,batch_size = 1,
|
103 |
+
num_inference_steps=20, s=1.5,
|
104 |
+
num_evolution_per_mask=1).images
|
105 |
+
|
106 |
+
return images
|
requirements.txt
ADDED
@@ -0,0 +1,137 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
accelerate==0.25.0
|
2 |
+
aiofiles==23.2.1
|
3 |
+
aiohttp==3.9.1
|
4 |
+
aiosignal==1.3.1
|
5 |
+
albumentations==1.3.1
|
6 |
+
altair==5.2.0
|
7 |
+
annotated-types==0.6.0
|
8 |
+
antlr4-python3-runtime==4.9.3
|
9 |
+
anyio==4.2.0
|
10 |
+
appdirs==1.4.4
|
11 |
+
async-timeout==4.0.3
|
12 |
+
attrs==23.2.0
|
13 |
+
blobfile==2.1.1
|
14 |
+
certifi==2023.11.17
|
15 |
+
charset-normalizer==3.3.2
|
16 |
+
click==8.1.7
|
17 |
+
colorama==0.4.6
|
18 |
+
contourpy==1.2.0
|
19 |
+
cycler==0.12.1
|
20 |
+
datasets==2.16.1
|
21 |
+
diffusers==0.16.1
|
22 |
+
dill==0.3.7
|
23 |
+
docker-pycreds==0.4.0
|
24 |
+
einops==0.7.0
|
25 |
+
exceptiongroup==1.2.0
|
26 |
+
fastapi==0.109.2
|
27 |
+
ffmpy==0.3.1
|
28 |
+
filelock==3.13.1
|
29 |
+
fonttools==4.47.0
|
30 |
+
frozenlist==1.4.1
|
31 |
+
fsspec==2023.10.0
|
32 |
+
gitdb==4.0.11
|
33 |
+
GitPython==3.1.41
|
34 |
+
gradio==4.16.0
|
35 |
+
gradio_client==0.8.1
|
36 |
+
h11==0.14.0
|
37 |
+
httpcore==1.0.2
|
38 |
+
httpx==0.26.0
|
39 |
+
huggingface-hub==0.20.1
|
40 |
+
idna==3.6
|
41 |
+
imageio==2.33.1
|
42 |
+
importlib-metadata==7.0.1
|
43 |
+
importlib-resources==6.1.1
|
44 |
+
Jinja2==3.1.2
|
45 |
+
joblib==1.3.2
|
46 |
+
jsonschema==4.21.1
|
47 |
+
jsonschema-specifications==2023.12.1
|
48 |
+
kiwisolver==1.4.5
|
49 |
+
lazy_loader==0.3
|
50 |
+
lightning-utilities==0.10.0
|
51 |
+
lxml==4.9.4
|
52 |
+
markdown-it-py==3.0.0
|
53 |
+
MarkupSafe==2.1.3
|
54 |
+
matplotlib==3.8.2
|
55 |
+
mdurl==0.1.2
|
56 |
+
mpmath==1.3.0
|
57 |
+
multidict==6.0.4
|
58 |
+
multiprocess==0.70.15
|
59 |
+
networkx==3.2.1
|
60 |
+
numpy==1.26.2
|
61 |
+
nvidia-cublas-cu12==12.1.3.1
|
62 |
+
nvidia-cuda-cupti-cu12==12.1.105
|
63 |
+
nvidia-cuda-nvrtc-cu12==12.1.105
|
64 |
+
nvidia-cuda-runtime-cu12==12.1.105
|
65 |
+
nvidia-cudnn-cu12==8.9.2.26
|
66 |
+
nvidia-cufft-cu12==11.0.2.54
|
67 |
+
nvidia-curand-cu12==10.3.2.106
|
68 |
+
nvidia-cusolver-cu12==11.4.5.107
|
69 |
+
nvidia-cusparse-cu12==12.1.0.106
|
70 |
+
nvidia-nccl-cu12==2.18.1
|
71 |
+
nvidia-nvjitlink-cu12==12.3.101
|
72 |
+
nvidia-nvtx-cu12==12.1.105
|
73 |
+
omegaconf==2.3.0
|
74 |
+
opencv-contrib-python==4.9.0.80
|
75 |
+
opencv-python-headless==4.9.0.80
|
76 |
+
orjson==3.9.13
|
77 |
+
packaging==23.2
|
78 |
+
pandas==2.1.4
|
79 |
+
pillow==10.2.0
|
80 |
+
protobuf==4.25.2
|
81 |
+
psutil==5.9.7
|
82 |
+
pyarrow==14.0.2
|
83 |
+
pyarrow-hotfix==0.6
|
84 |
+
pycryptodomex==3.20.0
|
85 |
+
pydantic==2.6.0
|
86 |
+
pydantic_core==2.16.1
|
87 |
+
pydub==0.25.1
|
88 |
+
Pygments==2.17.2
|
89 |
+
pyparsing==3.1.1
|
90 |
+
python-dateutil==2.8.2
|
91 |
+
python-multipart==0.0.7
|
92 |
+
pytorch-lightning==2.1.3
|
93 |
+
pytz==2023.3.post1
|
94 |
+
PyYAML==6.0.1
|
95 |
+
qudida==0.0.4
|
96 |
+
referencing==0.33.0
|
97 |
+
regex==2023.12.25
|
98 |
+
requests==2.31.0
|
99 |
+
rich==13.7.0
|
100 |
+
rpds-py==0.17.1
|
101 |
+
ruff==0.2.0
|
102 |
+
safetensors==0.4.1
|
103 |
+
scikit-image==0.22.0
|
104 |
+
scikit-learn==1.3.2
|
105 |
+
scipy==1.11.4
|
106 |
+
semantic-version==2.10.0
|
107 |
+
sentry-sdk==1.39.2
|
108 |
+
setproctitle==1.3.3
|
109 |
+
shellingham==1.5.4
|
110 |
+
six==1.16.0
|
111 |
+
smmap==5.0.1
|
112 |
+
sniffio==1.3.0
|
113 |
+
starlette==0.36.3
|
114 |
+
sympy==1.12
|
115 |
+
threadpoolctl==3.2.0
|
116 |
+
tifffile==2023.12.9
|
117 |
+
tokenizers==0.15.0
|
118 |
+
tomlkit==0.12.0
|
119 |
+
toolz==0.12.1
|
120 |
+
torch==2.1.2
|
121 |
+
torchaudio==2.1.2
|
122 |
+
torchmetrics==1.2.1
|
123 |
+
torchvision==0.16.2
|
124 |
+
tqdm==4.66.1
|
125 |
+
transformers==4.36.2
|
126 |
+
triton==2.1.0
|
127 |
+
typer==0.9.0
|
128 |
+
typing_extensions==4.9.0
|
129 |
+
tzdata==2023.4
|
130 |
+
urllib3==2.1.0
|
131 |
+
uvicorn==0.27.0.post1
|
132 |
+
wandb==0.16.2
|
133 |
+
websockets==11.0.3
|
134 |
+
xformers==0.0.23.post1
|
135 |
+
xxhash==3.4.1
|
136 |
+
yarl==1.9.4
|
137 |
+
zipp==3.17.0
|